JSON은 애플리케이션 간에 데이터를 교환할 때 전 세계적으로 가장 널리 쓰이는 형식 중 하나입니다.
구조화된 출력값은 모델이 항상 지정한 JSON 스키마에 맞춰 응답을 생성하도록 하는 기능입니다. 모델이 필수 키를 누락하거나 열거형에 정의되지 않은 값을 지어낼 걱정을 할 필요가 없습니다.
구조화된 출력값을 사용하면 다음과 같은 이점이 있습니다.
- 타입 안전성 보장: 응답 형식을 별도로 검증하거나 잘못된 형식 때문에 요청을 다시 보낼 필요가 없습니다.
- 명시적인 거부 응답: 모델이 안전상의 이유로 요청을 거부했는지 코드에서 확인할 수 있습니다.
- 간단한 프롬프트: 일관된 응답 형식을 얻기 위해 프롬프트에서 형식을 거듭 강조할 필요가 없습니다.
OpenAI의 REST API는 JSON 스키마를 지원합니다. Python과 JavaScript용 OpenAI 라이브러리에서는 각각 pydantic.BaseModel과 z.object로 객체 스키마를 정의할 수도 있습니다. 아래 예제에서는 비정형 텍스트에서 정보를 추출해 코드로 정의한 스키마에 맞춰 반환하는 방법을 보여 줍니다.
Ruby SDK는 Sorbet의 T::Struct로 정의한 스키마를 지원하며, 파싱한 결과를 해당 타입으로 반환합니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25import OpenAI from "openai";
import { zodResponseFormat } from "openai/helpers/zod";
import { z } from "zod";
const openai = new OpenAI();
const CalendarEvent = z.object({
name: z.string(),
date: z.string(),
participants: z.array(z.string()),
});
const completion = await openai.chat.completions.parse({
model: "gpt-6-astra",
messages: [
{ role: "system", content: "Extract the event information." },
{
role: "user",
content: "Alice and Bob are going to a science fair on Friday.",
},
],
response_format: zodResponseFormat(CalendarEvent, "event"),
});
const event = completion.choices[0].message.parsed;1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25from pydantic import BaseModel
from openai import OpenAI
client = OpenAI()
class CalendarEvent(BaseModel):
name: str
date: str
participants: list[str]
completion = client.chat.completions.parse(
model="gpt-6-astra",
messages=[
{"role": "system", "content": "Extract the event information."},
{
"role": "user",
"content": "Alice and Bob are going to a science fair on Friday.",
},
],
response_format=CalendarEvent,
)
event = completion.choices[0].message.parsed1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/shared"
)
func main() {
client := openai.NewClient()
schema := map[string]any{
"type": "object",
"properties": map[string]any{
"name": map[string]any{"type": "string"},
"date": map[string]any{"type": "string"},
"participants": map[string]any{"type": "array", "items": map[string]any{"type": "string"}},
},
"required": []string{"name", "date", "participants"},
"additionalProperties": false,
}
completion, err := client.Chat.Completions.New(context.Background(), openai.ChatCompletionNewParams{
Model: "gpt-6-astra",
Messages: []openai.ChatCompletionMessageParamUnion{
openai.SystemMessage("Extract the event information."),
openai.UserMessage("Alice and Bob are going to a science fair on Friday."),
},
ResponseFormat: openai.ChatCompletionNewParamsResponseFormatUnion{
OfJSONSchema: &shared.ResponseFormatJSONSchemaParam{JSONSchema: shared.ResponseFormatJSONSchemaJSONSchemaParam{
Name: "event", Schema: schema, Strict: openai.Bool(true),
}},
},
})
if err != nil {
panic(err)
}
fmt.Println(completion.Choices[0].Message.Content)
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.chat.completions.ChatCompletionCreateParams;
import java.util.List;
import java.util.Map;
Map<String, Object> schema =
Map.of(
"type",
"object",
"properties",
Map.of(
"name", Map.of("type", "string"),
"date", Map.of("type", "string"),
"participants", Map.of("type", "array", "items", Map.of("type", "string"))),
"required",
List.of("name", "date", "participants"),
"additionalProperties",
false);
ChatCompletionCreateParams params =
ChatCompletionCreateParams.builder()
.model("gpt-6-astra")
.addSystemMessage("Extract the event information.")
.addUserMessage("Alice and Bob are going to a science fair on Friday.")
.putAdditionalBodyProperty(
"response_format",
JsonValue.from(
Map.of(
"type",
"json_schema",
"json_schema",
Map.of("name", "event", "strict", true, "schema", schema))))
.build();
client.chat().completions().create(params).choices().stream()
.flatMap(choice -> choice.message().content().stream())
.forEach(System.out::println);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42using OpenAI.Chat;
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
string model = "gpt-6-astra";
ChatClient client = new(model, key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"name": { "type": "string" },
"date": { "type": "string" },
"participants": {
"type": "array",
"items": { "type": "string" }
}
},
"required": ["name", "date", "participants"],
"additionalProperties": false
}
"""
);
ChatCompletionOptions options = new()
{
ResponseFormat = ChatResponseFormat.CreateJsonSchemaFormat(
"event",
schema,
jsonSchemaIsStrict: true
),
};
ChatCompletion completion = await client.CompleteChatAsync(
[
new SystemChatMessage("Extract the event information."),
new UserChatMessage(
"Alice and Bob are going to a science fair on Friday."
),
],
options
);
Console.WriteLine(completion.Content[0].Text);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34# gem install openai sorbet-runtime
require "openai"
require "openai/helpers/sorbet"
class CalendarEvent < T::Struct
const :name, String
const :date, String
const :participants, T::Array[String]
end
client = OpenAI::Client.new
schema = OpenAI::StructuredOutput.from_sorbet(CalendarEvent)
completion = client.chat.completions.create(
model: "gpt-6-astra",
messages: [
{
role: :system,
content: "Extract the event information."
},
{
role: :user,
content: "Alice and Bob are going to a science fair on Friday."
}
],
response_format: schema
)
choice = completion.choices.fetch(0)
raise "Completion ended with reason: #{choice.finish_reason}" unless choice.finish_reason.to_s == "stop"
raise "The model refused the request" if choice.message.refusal
event = T.cast(choice.message.parsed, CalendarEvent)
puts(event.name, event.date, event.participants.join(", "))1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27import OpenAI from "openai";
import { zodTextFormat } from "openai/helpers/zod";
import { z } from "zod";
const openai = new OpenAI();
const CalendarEvent = z.object({
name: z.string(),
date: z.string(),
participants: z.array(z.string()),
});
const response = await openai.responses.parse({
model: "gpt-6-astra",
input: [
{ role: "system", content: "Extract the event information." },
{
role: "user",
content: "Alice and Bob are going to a science fair on Friday.",
},
],
text: {
format: zodTextFormat(CalendarEvent, "event"),
},
});
const event = response.output_parsed;1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25from openai import OpenAI
from pydantic import BaseModel
client = OpenAI()
class CalendarEvent(BaseModel):
name: str
date: str
participants: list[str]
response = client.responses.parse(
model="gpt-6-astra",
input=[
{"role": "system", "content": "Extract the event information."},
{
"role": "user",
"content": "Alice and Bob are going to a science fair on Friday.",
},
],
text_format=CalendarEvent,
)
event = response.output_parsed1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/responses"
)
func main() {
client := openai.NewClient()
schema := map[string]any{
"type": "object",
"properties": map[string]any{
"name": map[string]any{"type": "string"},
"date": map[string]any{"type": "string"},
"participants": map[string]any{"type": "array", "items": map[string]any{"type": "string"}},
},
"required": []string{"name", "date", "participants"},
"additionalProperties": false,
}
response, err := client.Responses.New(context.Background(), responses.ResponseNewParams{
Model: "gpt-6-astra",
Input: responses.ResponseNewParamsInputUnion{OfInputItemList: responses.ResponseInputParam{
responses.ResponseInputItemParamOfMessage(
responses.ResponseInputMessageContentListParam{responses.ResponseInputContentParamOfInputText("Extract the event information.")},
responses.EasyInputMessageRoleSystem,
),
responses.ResponseInputItemParamOfMessage(
responses.ResponseInputMessageContentListParam{responses.ResponseInputContentParamOfInputText("Alice and Bob are going to a science fair on Friday.")},
responses.EasyInputMessageRoleUser,
),
}},
Text: responses.ResponseTextConfigParam{Format: responses.ResponseFormatTextConfigUnionParam{
OfJSONSchema: &responses.ResponseFormatTextJSONSchemaConfigParam{Name: "event", Schema: schema, Strict: openai.Bool(true)},
}},
})
if err != nil {
panic(err)
}
fmt.Println(response.OutputText())
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.responses.EasyInputMessage;
import com.openai.models.responses.ResponseCreateParams;
import com.openai.models.responses.ResponseFormatTextJsonSchemaConfig;
import com.openai.models.responses.ResponseInputItem;
import com.openai.models.responses.ResponseTextConfig;
import java.util.List;
import java.util.Map;
Map<String, Object> schema =
Map.of(
"type",
"object",
"properties",
Map.of(
"name", Map.of("type", "string"),
"date", Map.of("type", "string"),
"participants", Map.of("type", "array", "items", Map.of("type", "string"))),
"required",
List.of("name", "date", "participants"),
"additionalProperties",
false);
ResponseCreateParams params =
ResponseCreateParams.builder()
.model("gpt-6-astra")
.inputOfResponse(
List.of(
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.SYSTEM)
.content("Extract the event information.")
.build()),
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.USER)
.content("Alice and Bob are going to a science fair on Friday.")
.build())))
.text(
ResponseTextConfig.builder()
.format(
ResponseFormatTextJsonSchemaConfig.builder()
.name("event")
.strict(true)
.schema(
JsonValue.from(schema)
.convert(ResponseFormatTextJsonSchemaConfig.Schema.class))
.build())
.build())
.build();
client.responses().create(params).output().stream()
.flatMap(item -> item.message().stream())
.flatMap(message -> message.content().stream())
.flatMap(content -> content.outputText().stream())
.forEach(text -> System.out.println(text.text()));1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47using OpenAI.Responses;
#pragma warning disable OPENAI001
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ResponsesClient client = new(key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"name": { "type": "string" },
"date": { "type": "string" },
"participants": {
"type": "array",
"items": { "type": "string" }
}
},
"required": ["name", "date", "participants"],
"additionalProperties": false
}
"""
);
CreateResponseOptions options = new()
{
Model = "gpt-6-astra",
TextOptions = new ResponseTextOptions
{
TextFormat = ResponseTextFormat.CreateJsonSchemaFormat(
"event",
schema,
jsonSchemaIsStrict: true
),
},
};
options.InputItems.Add(
ResponseItem.CreateSystemMessageItem("Extract the event information.")
);
options.InputItems.Add(
ResponseItem.CreateUserMessageItem(
"Alice and Bob are going to a science fair on Friday."
)
);
ResponseResult response = await client.CreateResponseAsync(options);
Console.WriteLine(response.GetOutputText());1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36# gem install openai sorbet-runtime
require "openai"
require "openai/helpers/sorbet"
class CalendarEvent < T::Struct
const :name, String
const :date, String
const :participants, T::Array[String]
end
client = OpenAI::Client.new
schema = OpenAI::StructuredOutput.from_sorbet(CalendarEvent)
response = client.responses.create(
model: "gpt-6-astra",
input: [
{
role: :system,
content: "Extract the event information."
},
{
role: :user,
content: "Alice and Bob are going to a science fair on Friday."
}
],
text: schema
)
raise "Response ended with status: #{response.status}" unless response.status == OpenAI::Responses::ResponseStatus::COMPLETED
message = response.output.grep(OpenAI::Responses::ResponseOutputMessage).fetch(0)
output_text = message.content.grep(OpenAI::Responses::ResponseOutputText).first
raise "No structured output returned (the model may have refused)" unless output_text
event = T.cast(output_text.parsed, CalendarEvent)
puts(event.name, event.date, event.participants.join(", "))지원 모델
GPT-4o부터 OpenAI의 최신 대규모 언어 모델에서 구조화된 출력값을 사용할 수 있습니다. 새 프로젝트에서는 gpt-6-astra로 시작하세요. gpt-4-turbo와 그 이전 모델에서는 대신 JSON 모드를 사용할 수 있습니다.
구조화된 출력값의 용도별 선택 기준: 함수 호출과 response_format
구조화된 출력값의 용도별 선택 기준: 함수 호출과 text.format
OpenAI API에서 구조화된 출력값을 사용하는 방법은 두 가지입니다.
- 모델의 함수 호출 사용
- 응답 형식으로
json_schema사용
함수 호출은 모델을 애플리케이션의 기능과 연결할 때 유용합니다.
예를 들어 모델이 데이터베이스를 조회하는 함수를 호출하도록 하면 사용자의 주문 관련 문의를 처리하는 AI 어시스턴트를 만들 수 있습니다. 모델에 UI와 상호작용하는 함수를 제공할 수도 있습니다.
반면 모델이 도구를 호출할 때가 아니라 사용자에게 응답할 때 따를 스키마를 지정하려면 response_format을 사용하는 구조화된 출력값이 더 적합합니다.
예를 들어 수학 학습 애플리케이션에서는 어시스턴트가 특정 JSON 스키마에 맞춰 응답하도록 할 수 있습니다. 그러면 모델 출력의 각 부분을 서로 다른 방식으로 표시하는 UI를 만들 수 있습니다.
용도에 따라 다음과 같이 선택하세요.
- 모델을 시스템의 도구, 함수, 데이터 등에 연결하려면 함수 호출을 사용하세요. 모델이 사용자에게 보내는 응답을 구조화하려면 구조화된
response_format을 사용하세요.
- 모델을 시스템의 도구, 함수, 데이터 등에 연결하려면 함수 호출을 사용하세요. 모델이 사용자에게 보내는 응답을 구조화하려면 구조화된
text.format을 사용하세요.
이 가이드의 나머지 부분에서는 Chat Completions API에서 함수 호출 없이 구조화된 출력값을 사용하는 사례를 다룹니다. 함수 호출과 함께 구조화된 출력값을 사용하는 방법은
함수 호출
가이드를 참고하세요.
이 가이드의 나머지 부분에서는 Responses API에서 함수 호출 없이 구조화된 출력값을 사용하는 사례를 다룹니다. 함수 호출과 함께 구조화된 출력값을 사용하는 방법은
함수 호출
가이드를 참고하세요.
구조화된 출력값과 JSON 모드 비교
구조화된 출력값은 JSON 모드를 발전시킨 기능입니다. 두 기능 모두 유효한 JSON을 생성하지만, 지정한 스키마까지 따르도록 보장하는 기능은 구조화된 출력값뿐입니다. Responses API, Chat Completions API, Assistants API, Fine-tuning API, Batch API에서 두 기능을 모두 지원합니다.
가능하면 항상 JSON 모드 대신 구조화된 출력값을 사용하는 것이 좋습니다.
다만 response_format: {type: "json_schema", ...}를 사용하는 구조화된 출력값은 gpt-4o-mini, gpt-4o-mini-2024-07-18, gpt-4o-2024-08-06 모델 스냅샷과 그 이후 버전에서만 지원됩니다.
| 구조화된 출력값 | JSON 모드 | |
|---|---|---|
| 유효한 JSON 출력 | 예 | 예 |
| 스키마 준수 | 예(지원되는 스키마 참고) | 아니요 |
| 호환 모델 | gpt-4o-mini, gpt-4o-2024-08-06 및 이후 모델 |
gpt-3.5-turbo, gpt-4-*, gpt-4o-* 및 호환되는 GPT-5 모델 |
| 활성화 방법 | response_format: { type: "json_schema", json_schema: {"strict": true, "schema": ...} } |
response_format: { type: "json_object" } |
| 구조화된 출력값 | JSON 모드 | |
|---|---|---|
| 유효한 JSON 출력 | 예 | 예 |
| 스키마 준수 | 예(지원되는 스키마 참고) | 아니요 |
| 호환 모델 | gpt-4o-mini, gpt-4o-2024-08-06 및 이후 모델 |
gpt-3.5-turbo, gpt-4-*, gpt-4o-* 및 호환되는 GPT-5 모델 |
| 활성화 방법 | text: { format: { type: "json_schema", "strict": true, "schema": ... } } |
text: { format: { type: "json_object" } } |
예제
추론 과정
사용자가 풀이 과정을 따라갈 수 있도록 모델에 답변을 단계별로 구조화해 출력하도록 요청할 수 있습니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30import OpenAI from "openai";
import { z } from "zod";
import { zodResponseFormat } from "openai/helpers/zod";
const openai = new OpenAI();
const Step = z.object({
explanation: z.string(),
output: z.string(),
});
const MathReasoning = z.object({
steps: z.array(Step),
final_answer: z.string(),
});
const completion = await openai.chat.completions.parse({
model: "gpt-6-astra",
messages: [
{
role: "system",
content:
"You are a helpful math tutor. Guide the user through the solution step by step.",
},
{ role: "user", content: "how can I solve 8x + 7 = -23" },
],
response_format: zodResponseFormat(MathReasoning, "math_reasoning"),
});
const math_reasoning = completion.choices[0].message.parsed;1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29from pydantic import BaseModel
from openai import OpenAI
client = OpenAI()
class Step(BaseModel):
explanation: str
output: str
class MathReasoning(BaseModel):
steps: list[Step]
final_answer: str
completion = client.chat.completions.parse(
model="gpt-6-astra",
messages=[
{
"role": "system",
"content": "You are a helpful math tutor. Guide the user through the solution step by step.",
},
{"role": "user", "content": "how can I solve 8x + 7 = -23"},
],
response_format=MathReasoning,
)
math_reasoning = completion.choices[0].message.parsed1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/shared"
)
func main() {
client := openai.NewClient()
step := map[string]any{
"type": "object",
"properties": map[string]any{
"explanation": map[string]any{"type": "string"},
"output": map[string]any{"type": "string"},
},
"required": []string{"explanation", "output"},
"additionalProperties": false,
}
schema := map[string]any{
"type": "object",
"properties": map[string]any{
"steps": map[string]any{"type": "array", "items": step},
"final_answer": map[string]any{"type": "string"},
},
"required": []string{"steps", "final_answer"},
"additionalProperties": false,
}
completion, err := client.Chat.Completions.New(context.Background(), openai.ChatCompletionNewParams{
Model: "gpt-6-astra",
Messages: []openai.ChatCompletionMessageParamUnion{
openai.SystemMessage("You are a helpful math tutor. Guide the user through the solution step by step."),
openai.UserMessage("how can I solve 8x + 7 = -23"),
},
ResponseFormat: openai.ChatCompletionNewParamsResponseFormatUnion{
OfJSONSchema: &shared.ResponseFormatJSONSchemaParam{JSONSchema: shared.ResponseFormatJSONSchemaJSONSchemaParam{
Name: "math_reasoning", Schema: schema, Strict: openai.Bool(true),
}},
},
})
if err != nil {
panic(err)
}
fmt.Println(completion.Choices[0].Message.Content)
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.chat.completions.ChatCompletionCreateParams;
import java.util.List;
import java.util.Map;
Map<String, Object> schema =
Map.of(
"type",
"object",
"properties",
Map.of(
"steps",
Map.of(
"type",
"array",
"items",
Map.of(
"type",
"object",
"properties",
Map.of(
"explanation", Map.of("type", "string"),
"output", Map.of("type", "string")),
"required",
List.of("explanation", "output"),
"additionalProperties",
false)),
"final_answer", Map.of("type", "string")),
"required",
List.of("steps", "final_answer"),
"additionalProperties",
false);
ChatCompletionCreateParams params =
ChatCompletionCreateParams.builder()
.model("gpt-6-astra")
.addSystemMessage(
"You are a helpful math tutor. Guide the user through the solution step by step.")
.addUserMessage("How can I solve 8x + 7 = -23?")
.putAdditionalBodyProperty(
"response_format",
JsonValue.from(
Map.of(
"type",
"json_schema",
"json_schema",
Map.of("name", "math_reasoning", "strict", true, "schema", schema))))
.build();
client.chat().completions().create(params).choices().stream()
.flatMap(choice -> choice.message().content().stream())
.forEach(System.out::println);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46using System.Text.Json;
using OpenAI.Chat;
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
string model = "gpt-6-astra";
ChatClient client = new(model, key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": { "type": "string" },
"output": { "type": "string" }
},
"required": ["explanation", "output"],
"additionalProperties": false
}
},
"final_answer": { "type": "string" }
},
"required": ["steps", "final_answer"],
"additionalProperties": false
}
"""
);
ChatCompletionOptions options = new()
{
ResponseFormat = ChatResponseFormat.CreateJsonSchemaFormat(
"math_response",
schema,
jsonSchemaIsStrict: true
),
};
ChatCompletion completion = await client.CompleteChatAsync(
[new SystemChatMessage("You are a helpful math tutor. Guide the user through the solution step by step."), new UserChatMessage("How can I solve 8x + 7 = -23?")],
options
);
using JsonDocument parsed = JsonDocument.Parse(completion.Content[0].Text);
Console.WriteLine(parsed.RootElement);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48require "openai"
client = OpenAI::Client.new
step_schema = {
type: :object,
properties: {
explanation: { type: :string },
output: { type: :string }
},
required: %w[explanation output],
additionalProperties: false
}
math_schema = {
type: :object,
properties: {
steps: {
type: :array,
items: step_schema
},
final_answer: { type: :string }
},
required: %w[steps final_answer],
additionalProperties: false
}
completion = client.chat.completions.create(
model: "gpt-6-astra",
messages: [
{
role: :system,
content: "You are a helpful math tutor. Guide the user through the solution step by step."
},
{
role: :user,
content: "How can I solve 8x + 7 = -23?"
}
],
response_format: {
type: :json_schema,
json_schema: {
name: "math_reasoning",
strict: true,
schema: math_schema
}
}
)
puts(completion.choices.fetch(0).message.content)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43curl https://api.openai.com/v1/chat/completions \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6-astra",
"messages": [
{
"role": "system",
"content": "You are a helpful math tutor. Guide the user through the solution step by step."
},
{
"role": "user",
"content": "how can I solve 8x + 7 = -23"
}
],
"response_format": {
"type": "json_schema",
"json_schema": {
"name": "math_reasoning",
"schema": {
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": { "type": "string" },
"output": { "type": "string" }
},
"required": ["explanation", "output"],
"additionalProperties": false
}
},
"final_answer": { "type": "string" }
},
"required": ["steps", "final_answer"],
"additionalProperties": false
},
"strict": true
}
}
}'1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32import OpenAI from "openai";
import { zodTextFormat } from "openai/helpers/zod";
import { z } from "zod";
const openai = new OpenAI();
const Step = z.object({
explanation: z.string(),
output: z.string(),
});
const MathReasoning = z.object({
steps: z.array(Step),
final_answer: z.string(),
});
const response = await openai.responses.parse({
model: "gpt-6-astra",
input: [
{
role: "system",
content:
"You are a helpful math tutor. Guide the user through the solution step by step.",
},
{ role: "user", content: "how can I solve 8x + 7 = -23" },
],
text: {
format: zodTextFormat(MathReasoning, "math_reasoning"),
},
});
const math_reasoning = response.output_parsed;1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29from openai import OpenAI
from pydantic import BaseModel
client = OpenAI()
class Step(BaseModel):
explanation: str
output: str
class MathReasoning(BaseModel):
steps: list[Step]
final_answer: str
response = client.responses.parse(
model="gpt-6-astra",
input=[
{
"role": "system",
"content": "You are a helpful math tutor. Guide the user through the solution step by step.",
},
{"role": "user", "content": "how can I solve 8x + 7 = -23"},
],
text_format=MathReasoning,
)
math_reasoning = response.output_parsed1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/responses"
)
func main() {
client := openai.NewClient()
step := map[string]any{
"type": "object",
"properties": map[string]any{
"explanation": map[string]any{"type": "string"},
"output": map[string]any{"type": "string"},
},
"required": []string{"explanation", "output"},
"additionalProperties": false,
}
schema := map[string]any{
"type": "object",
"properties": map[string]any{
"steps": map[string]any{"type": "array", "items": step},
"final_answer": map[string]any{"type": "string"},
},
"required": []string{"steps", "final_answer"},
"additionalProperties": false,
}
response, err := client.Responses.New(context.Background(), responses.ResponseNewParams{
Model: "gpt-6-astra",
Input: responses.ResponseNewParamsInputUnion{OfInputItemList: responses.ResponseInputParam{
responses.ResponseInputItemParamOfMessage(
responses.ResponseInputMessageContentListParam{responses.ResponseInputContentParamOfInputText("You are a helpful math tutor. Guide the user through the solution step by step.")},
responses.EasyInputMessageRoleSystem,
),
responses.ResponseInputItemParamOfMessage(
responses.ResponseInputMessageContentListParam{responses.ResponseInputContentParamOfInputText("how can I solve 8x + 7 = -23")},
responses.EasyInputMessageRoleUser,
),
}},
Text: responses.ResponseTextConfigParam{Format: responses.ResponseFormatTextConfigUnionParam{
OfJSONSchema: &responses.ResponseFormatTextJSONSchemaConfigParam{Name: "math_reasoning", Schema: schema, Strict: openai.Bool(true)},
}},
})
if err != nil {
panic(err)
}
fmt.Println(response.OutputText())
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.responses.EasyInputMessage;
import com.openai.models.responses.ResponseCreateParams;
import com.openai.models.responses.ResponseFormatTextJsonSchemaConfig;
import com.openai.models.responses.ResponseInputItem;
import com.openai.models.responses.ResponseTextConfig;
import java.util.List;
import java.util.Map;
Map<String, Object> schema =
Map.of(
"type",
"object",
"properties",
Map.of(
"steps",
Map.of(
"type",
"array",
"items",
Map.of(
"type",
"object",
"properties",
Map.of(
"explanation", Map.of("type", "string"),
"output", Map.of("type", "string")),
"required",
List.of("explanation", "output"),
"additionalProperties",
false)),
"final_answer", Map.of("type", "string")),
"required",
List.of("steps", "final_answer"),
"additionalProperties",
false);
ResponseCreateParams params =
ResponseCreateParams.builder()
.model("gpt-6-astra")
.inputOfResponse(
List.of(
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.SYSTEM)
.content(
"You are a helpful math tutor. Guide the user through the solution step by step.")
.build()),
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.USER)
.content("How can I solve 8x + 7 = -23?")
.build())))
.text(
ResponseTextConfig.builder()
.format(
ResponseFormatTextJsonSchemaConfig.builder()
.name("math_reasoning")
.strict(true)
.schema(
JsonValue.from(schema)
.convert(ResponseFormatTextJsonSchemaConfig.Schema.class))
.build())
.build())
.build();
client.responses().create(params).output().stream()
.flatMap(item -> item.message().stream())
.flatMap(message -> message.content().stream())
.flatMap(content -> content.outputText().stream())
.forEach(text -> System.out.println(text.text()));1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49using System.Text.Json;
using OpenAI.Responses;
#pragma warning disable OPENAI001
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ResponsesClient client = new(key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": { "type": "string" },
"output": { "type": "string" }
},
"required": ["explanation", "output"],
"additionalProperties": false
}
},
"final_answer": { "type": "string" }
},
"required": ["steps", "final_answer"],
"additionalProperties": false
}
"""
);
CreateResponseOptions options = new()
{
Model = "gpt-6-astra",
TextOptions = new ResponseTextOptions
{
TextFormat = ResponseTextFormat.CreateJsonSchemaFormat(
"math_response",
schema,
jsonSchemaIsStrict: true
),
},
};
options.InputItems.Add(ResponseItem.CreateSystemMessageItem("You are a helpful math tutor. Guide the user through the solution step by step."));
options.InputItems.Add(ResponseItem.CreateUserMessageItem("How can I solve 8x + 7 = -23?"));
ResponseResult response = await client.CreateResponseAsync(options);
using JsonDocument parsed = JsonDocument.Parse(response.GetOutputText());
Console.WriteLine(parsed.RootElement);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48require "openai"
client = OpenAI::Client.new
step_schema = {
type: :object,
properties: {
explanation: { type: :string },
output: { type: :string }
},
required: %w[explanation output],
additionalProperties: false
}
math_schema = {
type: :object,
properties: {
steps: {
type: :array,
items: step_schema
},
final_answer: { type: :string }
},
required: %w[steps final_answer],
additionalProperties: false
}
response = client.responses.create(
model: "gpt-6-astra",
input: [
{
role: :system,
content: "You are a helpful math tutor. Guide the user through the solution step by step."
},
{
role: :user,
content: "How can I solve 8x + 7 = -23?"
}
],
text: {
format: {
type: :json_schema,
name: "math_reasoning",
strict: true,
schema: math_schema
}
}
)
puts(response.output_text)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43curl https://api.openai.com/v1/responses \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6-astra",
"input": [
{
"role": "system",
"content": "You are a helpful math tutor. Guide the user through the solution step by step."
},
{
"role": "user",
"content": "how can I solve 8x + 7 = -23"
}
],
"text": {
"format": {
"type": "json_schema",
"name": "math_reasoning",
"schema": {
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": { "type": "string" },
"output": { "type": "string" }
},
"required": ["explanation", "output"],
"additionalProperties": false
}
},
"final_answer": { "type": "string" }
},
"required": ["steps", "final_answer"],
"additionalProperties": false
},
"strict": true
}
}
}'응답 예시
{
"steps": [
{
"explanation": "Start with the equation 8x + 7 = -23.",
"output": "8x + 7 = -23"
},
{
"explanation": "Subtract 7 from both sides to isolate the term with the variable.",
"output": "8x = -23 - 7"
},
{
"explanation": "Simplify the right side of the equation.",
"output": "8x = -30"
},
{
"explanation": "Divide both sides by 8 to solve for x.",
"output": "x = -30 / 8"
},
{
"explanation": "Simplify the fraction.",
"output": "x = -15 / 4"
}
],
"final_answer": "x = -15 / 4"
}구조화된 데이터 추출
연구 논문과 같은 비정형 입력 데이터에서 어떤 정보를 추출할지 필드로 정의할 수 있습니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30import OpenAI from "openai";
import { z } from "zod";
import { zodResponseFormat } from "openai/helpers/zod";
const openai = new OpenAI();
const ResearchPaperExtraction = z.object({
title: z.string(),
authors: z.array(z.string()),
abstract: z.string(),
keywords: z.array(z.string()),
});
const completion = await openai.chat.completions.parse({
model: "gpt-6-astra",
messages: [
{
role: "system",
content:
"You are an expert at structured data extraction. You will be given unstructured text from a research paper and should convert it into the given structure.",
},
{ role: "user", content: "..." },
],
response_format: zodResponseFormat(
ResearchPaperExtraction,
"research_paper_extraction"
),
});
const research_paper = completion.choices[0].message.parsed;1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36from pydantic import BaseModel
from openai import OpenAI
client = OpenAI()
class ResearchPaperExtraction(BaseModel):
title: str
authors: list[str]
abstract: str
keywords: list[str]
completion = client.chat.completions.parse(
model="gpt-6-astra",
messages=[
{
"role": "system",
"content": "You are an expert at structured data extraction. You will be given unstructured text from a research paper and should convert it into the given structure.",
},
{
"role": "user",
"content": (
"Attention Is All You Need by Ashish Vaswani, Noam Shazeer, "
"Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, "
"Łukasz Kaiser, and Illia Polosukhin. We propose the "
"Transformer, a sequence transduction architecture based "
"entirely on attention. Keywords: transformers, attention, "
"sequence transduction."
),
},
],
response_format=ResearchPaperExtraction,
)
research_paper = completion.choices[0].message.parsed1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/shared"
)
const researchPaperText = "Attention Is All You Need by Ashish Vaswani, Noam Shazeer, " +
"Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, " +
"Łukasz Kaiser, and Illia Polosukhin. We propose the Transformer, " +
"a sequence transduction architecture based entirely on attention. " +
"Keywords: transformers, attention, sequence transduction."
func main() {
client := openai.NewClient()
schema := map[string]any{
"type": "object",
"properties": map[string]any{
"title": map[string]any{"type": "string"},
"authors": map[string]any{"type": "array", "items": map[string]any{"type": "string"}},
"abstract": map[string]any{"type": "string"},
"keywords": map[string]any{"type": "array", "items": map[string]any{"type": "string"}},
},
"required": []string{"title", "authors", "abstract", "keywords"},
"additionalProperties": false,
}
completion, err := client.Chat.Completions.New(context.Background(), openai.ChatCompletionNewParams{
Model: "gpt-6-astra",
Messages: []openai.ChatCompletionMessageParamUnion{
openai.SystemMessage("You are an expert at structured data extraction. You will be given unstructured text from a research paper and should convert it into the given structure."),
openai.UserMessage(researchPaperText),
},
ResponseFormat: openai.ChatCompletionNewParamsResponseFormatUnion{
OfJSONSchema: &shared.ResponseFormatJSONSchemaParam{JSONSchema: shared.ResponseFormatJSONSchemaJSONSchemaParam{
Name: "research_paper_extraction", Schema: schema, Strict: openai.Bool(true),
}},
},
})
if err != nil {
panic(err)
}
fmt.Println(completion.Choices[0].Message.Content)
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.chat.completions.ChatCompletionCreateParams;
import java.util.List;
import java.util.Map;
Map<String, Object> schema =
Map.of(
"type",
"object",
"properties",
Map.of(
"title", Map.of("type", "string"),
"authors", Map.of("type", "array", "items", Map.of("type", "string")),
"abstract", Map.of("type", "string"),
"keywords", Map.of("type", "array", "items", Map.of("type", "string"))),
"required",
List.of("title", "authors", "abstract", "keywords"),
"additionalProperties",
false);
ChatCompletionCreateParams params =
ChatCompletionCreateParams.builder()
.model("gpt-6-astra")
.addSystemMessage(
"You are an expert at structured data extraction. You will be given unstructured"
+ " text from a research paper and should convert it into the given structure.")
.addUserMessage(
"Attention Is All You Need by Ashish Vaswani, Noam Shazeer, Niki Parmar,"
+ " Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, and"
+ " Illia Polosukhin."
+ " We propose the Transformer, a sequence transduction architecture based"
+ " entirely on attention. Keywords: transformers, attention, sequence"
+ " transduction.")
.putAdditionalBodyProperty(
"response_format",
JsonValue.from(
Map.of(
"type",
"json_schema",
"json_schema",
Map.of(
"name",
"research_paper_extraction",
"strict",
true,
"schema",
schema))))
.build();
client.chat().completions().create(params).choices().stream()
.flatMap(choice -> choice.message().content().stream())
.forEach(System.out::println);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49using System.Text.Json;
using OpenAI.Chat;
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
string model = "gpt-6-astra";
ChatClient client = new(model, key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"title": { "type": "string" },
"authors": { "type": "array", "items": { "type": "string" } },
"abstract": { "type": "string" },
"keywords": { "type": "array", "items": { "type": "string" } }
},
"required": ["title", "authors", "abstract", "keywords"],
"additionalProperties": false
}
"""
);
ChatCompletionOptions options = new()
{
ResponseFormat = ChatResponseFormat.CreateJsonSchemaFormat(
"research_paper",
schema,
jsonSchemaIsStrict: true
),
};
ChatCompletion completion = await client.CompleteChatAsync(
[
new SystemChatMessage("Extract the title, authors, abstract, and keywords from the research paper."),
new UserChatMessage(
"""
Attention Is All You Need by Ashish Vaswani, Noam Shazeer,
Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez,
Łukasz Kaiser, and Illia Polosukhin. We propose the
Transformer, a sequence transduction architecture based
entirely on attention. Keywords: transformers, attention,
sequence transduction.
"""
),
],
options
);
using JsonDocument parsed = JsonDocument.Parse(completion.Content[0].Text);
Console.WriteLine(parsed.RootElement);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51require "openai"
client = OpenAI::Client.new
research_paper = <<~TEXT
Attention Is All You Need by Ashish Vaswani, Noam Shazeer, Niki Parmar,
Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, and Illia
Polosukhin. We propose the Transformer, a sequence transduction architecture
based entirely on attention. Keywords: transformers, attention, sequence
transduction.
TEXT
paper_schema = {
type: :object,
properties: {
title: { type: :string },
authors: {
type: :array,
items: { type: :string }
},
abstract: { type: :string },
keywords: {
type: :array,
items: { type: :string }
}
},
required: %w[title authors abstract keywords],
additionalProperties: false
}
completion = client.chat.completions.create(
model: "gpt-6-astra",
messages: [
{
role: :system,
content: "Extract structured data from the supplied research paper text."
},
{
role: :user,
content: research_paper
}
],
response_format: {
type: :json_schema,
json_schema: {
name: "research_paper_extraction",
strict: true,
schema: paper_schema
}
}
)
puts(completion.choices.fetch(0).message.content)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40curl https://api.openai.com/v1/chat/completions \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6-astra",
"messages": [
{
"role": "system",
"content": "You are an expert at structured data extraction. You will be given unstructured text from a research paper and should convert it into the given structure."
},
{
"role": "user",
"content": "..."
}
],
"response_format": {
"type": "json_schema",
"json_schema": {
"name": "research_paper_extraction",
"schema": {
"type": "object",
"properties": {
"title": { "type": "string" },
"authors": {
"type": "array",
"items": { "type": "string" }
},
"abstract": { "type": "string" },
"keywords": {
"type": "array",
"items": { "type": "string" }
}
},
"required": ["title", "authors", "abstract", "keywords"],
"additionalProperties": false
},
"strict": true
}
}
}'1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29import OpenAI from "openai";
import { zodTextFormat } from "openai/helpers/zod";
import { z } from "zod";
const openai = new OpenAI();
const ResearchPaperExtraction = z.object({
title: z.string(),
authors: z.array(z.string()),
abstract: z.string(),
keywords: z.array(z.string()),
});
const response = await openai.responses.parse({
model: "gpt-6-astra",
input: [
{
role: "system",
content:
"You are an expert at structured data extraction. You will be given unstructured text from a research paper and should convert it into the given structure.",
},
{ role: "user", content: "..." },
],
text: {
format: zodTextFormat(ResearchPaperExtraction, "research_paper_extraction"),
},
});
const research_paper = response.output_parsed;1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36from openai import OpenAI
from pydantic import BaseModel
client = OpenAI()
class ResearchPaperExtraction(BaseModel):
title: str
authors: list[str]
abstract: str
keywords: list[str]
response = client.responses.parse(
model="gpt-6-astra",
input=[
{
"role": "system",
"content": "You are an expert at structured data extraction. You will be given unstructured text from a research paper and should convert it into the given structure.",
},
{
"role": "user",
"content": (
"Attention Is All You Need by Ashish Vaswani, Noam Shazeer, "
"Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, "
"Łukasz Kaiser, and Illia Polosukhin. We propose the "
"Transformer, a sequence transduction architecture based "
"entirely on attention. Keywords: transformers, attention, "
"sequence transduction."
),
},
],
text_format=ResearchPaperExtraction,
)
research_paper = response.output_parsed1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/responses"
)
const researchPaperText = "Attention Is All You Need by Ashish Vaswani, Noam Shazeer, " +
"Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, " +
"Łukasz Kaiser, and Illia Polosukhin. We propose the Transformer, " +
"a sequence transduction architecture based entirely on attention. " +
"Keywords: transformers, attention, sequence transduction."
func main() {
client := openai.NewClient()
schema := map[string]any{
"type": "object",
"properties": map[string]any{
"title": map[string]any{"type": "string"},
"authors": map[string]any{"type": "array", "items": map[string]any{"type": "string"}},
"abstract": map[string]any{"type": "string"},
"keywords": map[string]any{"type": "array", "items": map[string]any{"type": "string"}},
},
"required": []string{"title", "authors", "abstract", "keywords"},
"additionalProperties": false,
}
response, err := client.Responses.New(context.Background(), responses.ResponseNewParams{
Model: "gpt-6-astra",
Input: responses.ResponseNewParamsInputUnion{OfInputItemList: responses.ResponseInputParam{
responses.ResponseInputItemParamOfMessage(
responses.ResponseInputMessageContentListParam{responses.ResponseInputContentParamOfInputText("You are an expert at structured data extraction. You will be given unstructured text from a research paper and should convert it into the given structure.")},
responses.EasyInputMessageRoleSystem,
),
responses.ResponseInputItemParamOfMessage(
responses.ResponseInputMessageContentListParam{responses.ResponseInputContentParamOfInputText(researchPaperText)},
responses.EasyInputMessageRoleUser,
),
}},
Text: responses.ResponseTextConfigParam{Format: responses.ResponseFormatTextConfigUnionParam{
OfJSONSchema: &responses.ResponseFormatTextJSONSchemaConfigParam{Name: "research_paper_extraction", Schema: schema, Strict: openai.Bool(true)},
}},
})
if err != nil {
panic(err)
}
fmt.Println(response.OutputText())
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.responses.EasyInputMessage;
import com.openai.models.responses.ResponseCreateParams;
import com.openai.models.responses.ResponseFormatTextJsonSchemaConfig;
import com.openai.models.responses.ResponseInputItem;
import com.openai.models.responses.ResponseTextConfig;
import java.util.List;
import java.util.Map;
Map<String, Object> schema =
Map.of(
"type",
"object",
"properties",
Map.of(
"title", Map.of("type", "string"),
"authors", Map.of("type", "array", "items", Map.of("type", "string")),
"abstract", Map.of("type", "string"),
"keywords", Map.of("type", "array", "items", Map.of("type", "string"))),
"required",
List.of("title", "authors", "abstract", "keywords"),
"additionalProperties",
false);
ResponseCreateParams params =
ResponseCreateParams.builder()
.model("gpt-6-astra")
.inputOfResponse(
List.of(
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.SYSTEM)
.content(
"You are an expert at structured data extraction. You will be given"
+ " unstructured text from a research paper and should convert"
+ " it into the given structure.")
.build()),
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.USER)
.content(
"Attention Is All You Need by Ashish Vaswani, Noam Shazeer,"
+ " Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez,"
+ " Łukasz Kaiser, and Illia Polosukhin. We propose the"
+ " Transformer, a"
+ " sequence transduction architecture based entirely on"
+ " attention. Keywords: transformers, attention, sequence"
+ " transduction.")
.build())))
.text(
ResponseTextConfig.builder()
.format(
ResponseFormatTextJsonSchemaConfig.builder()
.name("research_paper_extraction")
.strict(true)
.schema(
JsonValue.from(schema)
.convert(ResponseFormatTextJsonSchemaConfig.Schema.class))
.build())
.build())
.build();
client.responses().create(params).output().stream()
.flatMap(item -> item.message().stream())
.flatMap(message -> message.content().stream())
.flatMap(content -> content.outputText().stream())
.forEach(text -> System.out.println(text.text()));1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51using System.Text.Json;
using OpenAI.Responses;
#pragma warning disable OPENAI001
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ResponsesClient client = new(key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"title": { "type": "string" },
"authors": { "type": "array", "items": { "type": "string" } },
"abstract": { "type": "string" },
"keywords": { "type": "array", "items": { "type": "string" } }
},
"required": ["title", "authors", "abstract", "keywords"],
"additionalProperties": false
}
"""
);
CreateResponseOptions options = new()
{
Model = "gpt-6-astra",
TextOptions = new ResponseTextOptions
{
TextFormat = ResponseTextFormat.CreateJsonSchemaFormat(
"research_paper",
schema,
jsonSchemaIsStrict: true
),
},
};
options.InputItems.Add(ResponseItem.CreateSystemMessageItem("Extract the title, authors, abstract, and keywords from the research paper."));
options.InputItems.Add(
ResponseItem.CreateUserMessageItem(
"""
Attention Is All You Need by Ashish Vaswani, Noam Shazeer,
Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez,
Łukasz Kaiser, and Illia Polosukhin. We propose the
Transformer, a sequence transduction architecture based
entirely on attention. Keywords: transformers, attention,
sequence transduction.
"""
)
);
ResponseResult response = await client.CreateResponseAsync(options);
using JsonDocument parsed = JsonDocument.Parse(response.GetOutputText());
Console.WriteLine(parsed.RootElement);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51require "openai"
client = OpenAI::Client.new
research_paper = <<~TEXT
Attention Is All You Need by Ashish Vaswani, Noam Shazeer, Niki Parmar,
Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, and Illia
Polosukhin. We propose the Transformer, a sequence transduction architecture
based entirely on attention. Keywords: transformers, attention, sequence
transduction.
TEXT
paper_schema = {
type: :object,
properties: {
title: { type: :string },
authors: {
type: :array,
items: { type: :string }
},
abstract: { type: :string },
keywords: {
type: :array,
items: { type: :string }
}
},
required: %w[title authors abstract keywords],
additionalProperties: false
}
response = client.responses.create(
model: "gpt-6-astra",
input: [
{
role: :system,
content: "Extract structured data from the supplied research paper text."
},
{
role: :user,
content: research_paper
}
],
text: {
format: {
type: :json_schema,
name: "research_paper_extraction",
strict: true,
schema: paper_schema
}
}
)
puts(response.output_text)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40curl https://api.openai.com/v1/responses \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6-astra",
"input": [
{
"role": "system",
"content": "You are an expert at structured data extraction. You will be given unstructured text from a research paper and should convert it into the given structure."
},
{
"role": "user",
"content": "..."
}
],
"text": {
"format": {
"type": "json_schema",
"name": "research_paper_extraction",
"schema": {
"type": "object",
"properties": {
"title": { "type": "string" },
"authors": {
"type": "array",
"items": { "type": "string" }
},
"abstract": { "type": "string" },
"keywords": {
"type": "array",
"items": { "type": "string" }
}
},
"required": ["title", "authors", "abstract", "keywords"],
"additionalProperties": false
},
"strict": true
}
}
}'응답 예시
{
"title": "Application of Quantum Algorithms in Interstellar Navigation: A New Frontier",
"authors": ["Dr. Stella Voyager", "Dr. Nova Star", "Dr. Lyra Hunter"],
"abstract": "This paper investigates the utilization of quantum algorithms to improve interstellar navigation systems. By leveraging quantum superposition and entanglement, our proposed navigation system can calculate optimal travel paths through space-time anomalies more efficiently than classical methods. Experimental simulations suggest a significant reduction in travel time and fuel consumption for interstellar missions.",
"keywords": [
"Quantum algorithms",
"interstellar navigation",
"space-time anomalies",
"quantum superposition",
"quantum entanglement",
"space travel"
]
}UI 생성
HTML을 재귀적 데이터 구조로 표현하고 열거형 같은 제약 조건을 적용하면 유효한 HTML을 생성할 수 있습니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33import OpenAI from "openai";
import { z } from "zod";
import { zodResponseFormat } from "openai/helpers/zod";
const openai = new OpenAI();
const UI = z.lazy(() =>
z.object({
type: z.enum(["div", "button", "header", "section", "field", "form"]),
label: z.string(),
children: z.array(UI),
attributes: z.array(
z.object({
name: z.string(),
value: z.string(),
})
),
})
);
const completion = await openai.chat.completions.parse({
model: "gpt-6-astra",
messages: [
{
role: "system",
content: "You are a UI generator AI. Convert the user input into a UI.",
},
{ role: "user", content: "Make a User Profile Form" },
],
response_format: zodResponseFormat(UI, "ui"),
});
const ui = completion.choices[0].message.parsed;1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49from enum import Enum
from pydantic import BaseModel
from openai import OpenAI
client = OpenAI()
class UIType(str, Enum):
div = "div"
button = "button"
header = "header"
section = "section"
field = "field"
form = "form"
class Attribute(BaseModel):
name: str
value: str
class UI(BaseModel):
type: UIType
label: str
children: list["UI"]
attributes: list[Attribute]
UI.model_rebuild() # This is required to enable recursive types
class Response(BaseModel):
ui: UI
completion = client.chat.completions.parse(
model="gpt-6-astra",
messages=[
{
"role": "system",
"content": "You are a UI generator AI. Convert the user input into a UI.",
},
{"role": "user", "content": "Make a User Profile Form"},
],
response_format=Response,
)
ui = completion.choices[0].message.parsed
print(ui)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/shared"
)
func main() {
client := openai.NewClient()
schema := map[string]any{
"type": "object",
"properties": map[string]any{
"type": map[string]any{"type": "string", "enum": []string{"div", "button", "header", "section", "field", "form"}},
"label": map[string]any{"type": "string"},
"children": map[string]any{"type": "array", "items": map[string]any{"$ref": "#"}},
"attributes": map[string]any{"type": "array", "items": map[string]any{"type": "object", "properties": map[string]any{"name": map[string]any{"type": "string"}, "value": map[string]any{"type": "string"}}, "required": []string{"name", "value"}, "additionalProperties": false}},
},
"required": []string{"type", "label", "children", "attributes"},
"additionalProperties": false,
}
completion, err := client.Chat.Completions.New(context.Background(), openai.ChatCompletionNewParams{
Model: "gpt-6-astra",
Messages: []openai.ChatCompletionMessageParamUnion{
openai.SystemMessage("You are a UI generator AI. Convert the user input into a UI."),
openai.UserMessage("Make a User Profile Form"),
},
ResponseFormat: openai.ChatCompletionNewParamsResponseFormatUnion{
OfJSONSchema: &shared.ResponseFormatJSONSchemaParam{JSONSchema: shared.ResponseFormatJSONSchemaJSONSchemaParam{
Name: "ui", Description: openai.String("Dynamically generated UI"), Schema: schema, Strict: openai.Bool(true),
}},
},
})
if err != nil {
panic(err)
}
fmt.Println(completion.Choices[0].Message.Content)
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.chat.completions.ChatCompletion;
import com.openai.models.chat.completions.ChatCompletionCreateParams;
import java.util.LinkedHashMap;
import java.util.List;
import java.util.Map;
Map<String, Object> attributeProperties = new LinkedHashMap<>();
attributeProperties.put("name", Map.of("type", "string"));
attributeProperties.put("value", Map.of("type", "string"));
// Structured Outputs preserves the order of the schema's properties.
Map<String, Object> properties = new LinkedHashMap<>();
properties.put(
"type",
Map.of(
"type",
"string",
"enum",
List.of("div", "button", "header", "section", "field", "form")));
properties.put("label", Map.of("type", "string"));
properties.put("children", Map.of("type", "array", "items", Map.of("$ref", "#")));
properties.put(
"attributes",
Map.of(
"type",
"array",
"items",
Map.of(
"type",
"object",
"properties",
attributeProperties,
"required",
List.of("name", "value"),
"additionalProperties",
false)));
Map<String, Object> schema =
Map.of(
"type",
"object",
"properties",
properties,
"required",
List.of("type", "label", "children", "attributes"),
"additionalProperties",
false);
ChatCompletionCreateParams params =
ChatCompletionCreateParams.builder()
.model("gpt-6-astra")
.maxCompletionTokens(4096)
.addSystemMessage("Convert the user request into a UI definition.")
.addUserMessage("Make a user profile form.")
.putAdditionalBodyProperty(
"response_format",
JsonValue.from(
Map.of(
"type",
"json_schema",
"json_schema",
Map.of(
"name",
"ui",
"description",
"A dynamically generated UI",
"strict",
true,
"schema",
schema))))
.build();
var completion = client.chat().completions().create(params);
if (completion.choices().size() != 1) {
throw new IllegalStateException("UI generation did not complete.");
}
var choice = completion.choices().get(0);
if (!choice.finishReason().equals(ChatCompletion.Choice.FinishReason.STOP)
|| choice.message().refusal().isPresent()
|| choice.message().content().filter(text -> !text.isBlank()).isEmpty()) {
throw new IllegalStateException("UI generation did not complete.");
}
System.out.println(choice.message().content().orElseThrow());1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67using System.Text.Json;
using OpenAI.Chat;
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
string model = "gpt-6-astra";
ChatClient client = new(model, key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"ui": { "$ref": "#/$defs/component" }
},
"required": ["ui"],
"additionalProperties": false,
"$defs": {
"component": {
"type": "object",
"properties": {
"type": { "type": "string", "enum": ["div", "button", "header", "section", "field", "form"] },
"label": { "type": "string" },
"attributes": {
"type": "array",
"items": {
"type": "object",
"properties": { "name": { "type": "string" }, "value": { "type": "string" } },
"required": ["name", "value"],
"additionalProperties": false
}
},
"children": { "type": "array", "items": { "$ref": "#/$defs/component" } }
},
"required": ["type", "label", "children", "attributes"],
"additionalProperties": false
}
}
}
"""
);
ChatCompletionOptions options = new()
{
MaxOutputTokenCount = 4096,
ResponseFormat = ChatResponseFormat.CreateJsonSchemaFormat(
"ui",
schema,
jsonSchemaIsStrict: true
),
};
ChatCompletion completion = await client.CompleteChatAsync(
[
new SystemChatMessage("You are a UI generator. Convert the user request into a component tree."),
new UserChatMessage(
"Make a small user profile form with exactly three direct children: " +
"a Name field, an Email field, and a Save button. Give each child no children. " +
"Use short labels and at most two short attributes per component."
),
],
options
);
if (completion.FinishReason != ChatFinishReason.Stop || !string.IsNullOrEmpty(completion.Refusal) ||
completion.Content.Count != 1 || string.IsNullOrWhiteSpace(completion.Content[0].Text))
throw new InvalidOperationException("UI generation did not complete.");
using JsonDocument parsed = JsonDocument.Parse(completion.Content[0].Text);
Console.WriteLine(parsed.RootElement);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56require "openai"
client = OpenAI::Client.new
ui_schema = {
type: :object,
properties: {
type: {
type: :string,
enum: %w[div button header section field form]
},
label: { type: :string },
children: {
type: :array,
items: { "$ref" => "#" }
},
attributes: {
type: :array,
items: {
type: :object,
properties: {
name: { type: :string },
value: { type: :string }
},
required: %w[name value],
additionalProperties: false
}
}
},
required: %w[type label children attributes],
additionalProperties: false
}
completion = client.chat.completions.create(
model: "gpt-6-astra",
messages: [
{
role: :system,
content: "Convert the user request into a UI definition."
},
{
role: :user,
content: "Make a user profile form."
}
],
response_format: {
type: :json_schema,
json_schema: {
name: "ui",
description: "A dynamically generated UI",
strict: true,
schema: ui_schema
}
}
)
puts(completion.choices.fetch(0).message.content)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64curl https://api.openai.com/v1/chat/completions \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6-astra",
"messages": [
{
"role": "system",
"content": "You are a UI generator AI. Convert the user input into a UI."
},
{
"role": "user",
"content": "Make a User Profile Form"
}
],
"response_format": {
"type": "json_schema",
"json_schema": {
"name": "ui",
"description": "Dynamically generated UI",
"schema": {
"type": "object",
"properties": {
"type": {
"type": "string",
"description": "The type of the UI component",
"enum": ["div", "button", "header", "section", "field", "form"]
},
"label": {
"type": "string",
"description": "The label of the UI component, used for buttons or form fields"
},
"children": {
"type": "array",
"description": "Nested UI components",
"items": {"$ref": "#"}
},
"attributes": {
"type": "array",
"description": "Arbitrary attributes for the UI component, suitable for any element",
"items": {
"type": "object",
"properties": {
"name": {
"type": "string",
"description": "The name of the attribute, for example onClick or className"
},
"value": {
"type": "string",
"description": "The value of the attribute"
}
},
"required": ["name", "value"],
"additionalProperties": false
}
}
},
"required": ["type", "label", "children", "attributes"],
"additionalProperties": false
},
"strict": true
}
}
}'1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38import OpenAI from "openai";
import { zodTextFormat } from "openai/helpers/zod";
import { z } from "zod";
const openai = new OpenAI();
const UI = z.lazy(() =>
z.object({
type: z.enum(["div", "button", "header", "section", "field", "form"]),
label: z.string(),
children: z.array(UI),
attributes: z.array(
z.object({
name: z.string(),
value: z.string(),
})
),
})
);
const response = await openai.responses.parse({
model: "gpt-6-astra",
input: [
{
role: "system",
content: "You are a UI generator AI. Convert the user input into a UI.",
},
{
role: "user",
content: "Make a User Profile Form",
},
],
text: {
format: zodTextFormat(UI, "ui"),
},
});
const ui = response.output_parsed;1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49from enum import Enum
from openai import OpenAI
from pydantic import BaseModel
client = OpenAI()
class UIType(str, Enum):
div = "div"
button = "button"
header = "header"
section = "section"
field = "field"
form = "form"
class Attribute(BaseModel):
name: str
value: str
class UI(BaseModel):
type: UIType
label: str
children: list["UI"]
attributes: list[Attribute]
UI.model_rebuild() # This is required to enable recursive types
class Response(BaseModel):
ui: UI
response = client.responses.parse(
model="gpt-6-astra",
input=[
{
"role": "system",
"content": "You are a UI generator AI. Convert the user input into a UI.",
},
{"role": "user", "content": "Make a User Profile Form"},
],
text_format=Response,
)
ui = response.output_parsed1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/responses"
)
func main() {
client := openai.NewClient()
schema := map[string]any{
"type": "object",
"properties": map[string]any{
"type": map[string]any{"type": "string", "enum": []string{"div", "button", "header", "section", "field", "form"}},
"label": map[string]any{"type": "string"},
"children": map[string]any{"type": "array", "items": map[string]any{"$ref": "#"}},
"attributes": map[string]any{"type": "array", "items": map[string]any{"type": "object", "properties": map[string]any{"name": map[string]any{"type": "string"}, "value": map[string]any{"type": "string"}}, "required": []string{"name", "value"}, "additionalProperties": false}},
},
"required": []string{"type", "label", "children", "attributes"},
"additionalProperties": false,
}
response, err := client.Responses.New(context.Background(), responses.ResponseNewParams{
Model: "gpt-6-astra",
Input: responses.ResponseNewParamsInputUnion{OfInputItemList: responses.ResponseInputParam{
responses.ResponseInputItemParamOfMessage(
responses.ResponseInputMessageContentListParam{responses.ResponseInputContentParamOfInputText("You are a UI generator AI. Convert the user input into a UI.")},
responses.EasyInputMessageRoleSystem,
),
responses.ResponseInputItemParamOfMessage(
responses.ResponseInputMessageContentListParam{responses.ResponseInputContentParamOfInputText("Make a User Profile Form")},
responses.EasyInputMessageRoleUser,
),
}},
Text: responses.ResponseTextConfigParam{Format: responses.ResponseFormatTextConfigUnionParam{
OfJSONSchema: &responses.ResponseFormatTextJSONSchemaConfigParam{Name: "ui", Description: openai.String("Dynamically generated UI"), Schema: schema, Strict: openai.Bool(true)},
}},
})
if err != nil {
panic(err)
}
fmt.Println(response.OutputText())
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.responses.EasyInputMessage;
import com.openai.models.responses.ResponseCreateParams;
import com.openai.models.responses.ResponseFormatTextJsonSchemaConfig;
import com.openai.models.responses.ResponseInputItem;
import com.openai.models.responses.ResponseTextConfig;
import java.util.LinkedHashMap;
import java.util.List;
import java.util.Map;
Map<String, Object> attributeProperties = new LinkedHashMap<>();
attributeProperties.put("name", Map.of("type", "string"));
attributeProperties.put("value", Map.of("type", "string"));
// Structured Outputs preserves the order of the schema's properties.
Map<String, Object> properties = new LinkedHashMap<>();
properties.put(
"type",
Map.of(
"type",
"string",
"enum",
List.of("div", "button", "header", "section", "field", "form")));
properties.put("label", Map.of("type", "string"));
properties.put("children", Map.of("type", "array", "items", Map.of("$ref", "#")));
properties.put(
"attributes",
Map.of(
"type",
"array",
"items",
Map.of(
"type",
"object",
"properties",
attributeProperties,
"required",
List.of("name", "value"),
"additionalProperties",
false)));
ResponseCreateParams params =
ResponseCreateParams.builder()
.model("gpt-6-astra")
.inputOfResponse(
List.of(
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.SYSTEM)
.content("Convert the user request into a UI definition.")
.build()),
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.USER)
.content("Make a user profile form.")
.build())))
.text(
ResponseTextConfig.builder()
.format(
ResponseFormatTextJsonSchemaConfig.builder()
.name("ui")
.description("A dynamically generated UI")
.strict(true)
.schema(
ResponseFormatTextJsonSchemaConfig.Schema.builder()
.putAdditionalProperty("type", JsonValue.from("object"))
.putAdditionalProperty("properties", JsonValue.from(properties))
.putAdditionalProperty(
"required",
JsonValue.from(
List.of("type", "label", "children", "attributes")))
.putAdditionalProperty(
"additionalProperties", JsonValue.from(false))
.build())
.build())
.build())
.build();
client.responses().create(params).output().stream()
.flatMap(item -> item.message().stream())
.flatMap(message -> message.content().stream())
.flatMap(content -> content.outputText().stream())
.forEach(text -> System.out.println(text.text()));1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71using System.Text.Json;
using OpenAI.Responses;
#pragma warning disable OPENAI001
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ResponsesClient client = new(key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"ui": { "$ref": "#/$defs/component" }
},
"required": ["ui"],
"additionalProperties": false,
"$defs": {
"component": {
"type": "object",
"properties": {
"type": { "type": "string", "enum": ["div", "button", "header", "section", "field", "form"] },
"label": { "type": "string" },
"attributes": {
"type": "array",
"items": {
"type": "object",
"properties": { "name": { "type": "string" }, "value": { "type": "string" } },
"required": ["name", "value"],
"additionalProperties": false
}
},
"children": { "type": "array", "items": { "$ref": "#/$defs/component" } }
},
"required": ["type", "label", "children", "attributes"],
"additionalProperties": false
}
}
}
"""
);
CreateResponseOptions options = new()
{
Model = "gpt-6-astra",
MaxOutputTokenCount = 4096,
TextOptions = new ResponseTextOptions
{
TextFormat = ResponseTextFormat.CreateJsonSchemaFormat(
"ui",
schema,
jsonSchemaIsStrict: true
),
},
};
options.InputItems.Add(ResponseItem.CreateSystemMessageItem("You are a UI generator. Convert the user request into a component tree."));
options.InputItems.Add(ResponseItem.CreateUserMessageItem(
"Make a small user profile form with exactly three direct children: " +
"a Name field, an Email field, and a Save button. Give each child no children. " +
"Use short labels and at most two short attributes per component."
));
ResponseResult response = await client.CreateResponseAsync(options);
string text = response.GetOutputText();
bool refused = response.OutputItems.OfType<MessageResponseItem>()
.SelectMany(message => message.Content)
.Any(part => part.Kind == ResponseContentPartKind.Refusal);
if (response.Status != ResponseStatus.Completed || refused || string.IsNullOrWhiteSpace(text))
{
throw new InvalidOperationException("UI generation did not complete.");
}
using JsonDocument parsed = JsonDocument.Parse(text);
Console.WriteLine(parsed.RootElement);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56require "openai"
client = OpenAI::Client.new
ui_schema = {
type: :object,
properties: {
type: {
type: :string,
enum: %w[div button header section field form]
},
label: { type: :string },
children: {
type: :array,
items: { "$ref" => "#" }
},
attributes: {
type: :array,
items: {
type: :object,
properties: {
name: { type: :string },
value: { type: :string }
},
required: %w[name value],
additionalProperties: false
}
}
},
required: %w[type label children attributes],
additionalProperties: false
}
response = client.responses.create(
model: "gpt-6-astra",
input: [
{
role: :system,
content: "Convert the user request into a UI definition."
},
{
role: :user,
content: "Make a user profile form."
}
],
text: {
format: {
type: :json_schema,
name: "ui",
description: "A dynamically generated UI",
strict: true,
schema: ui_schema
}
}
)
puts(response.output_text)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64curl https://api.openai.com/v1/responses \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6-astra",
"input": [
{
"role": "system",
"content": "You are a UI generator AI. Convert the user input into a UI."
},
{
"role": "user",
"content": "Make a User Profile Form"
}
],
"text": {
"format": {
"type": "json_schema",
"name": "ui",
"description": "Dynamically generated UI",
"schema": {
"type": "object",
"properties": {
"type": {
"type": "string",
"description": "The type of the UI component",
"enum": ["div", "button", "header", "section", "field", "form"]
},
"label": {
"type": "string",
"description": "The label of the UI component, used for buttons or form fields"
},
"children": {
"type": "array",
"description": "Nested UI components",
"items": {"$ref": "#"}
},
"attributes": {
"type": "array",
"description": "Arbitrary attributes for the UI component, suitable for any element",
"items": {
"type": "object",
"properties": {
"name": {
"type": "string",
"description": "The name of the attribute, for example onClick or className"
},
"value": {
"type": "string",
"description": "The value of the attribute"
}
},
"required": ["name", "value"],
"additionalProperties": false
}
}
},
"required": ["type", "label", "children", "attributes"],
"additionalProperties": false
},
"strict": true
}
}
}'응답 예시
{
"type": "form",
"label": "User Profile Form",
"children": [
{
"type": "div",
"label": "",
"children": [
{
"type": "field",
"label": "First Name",
"children": [],
"attributes": [
{
"name": "type",
"value": "text"
},
{
"name": "name",
"value": "firstName"
},
{
"name": "placeholder",
"value": "Enter your first name"
}
]
},
{
"type": "field",
"label": "Last Name",
"children": [],
"attributes": [
{
"name": "type",
"value": "text"
},
{
"name": "name",
"value": "lastName"
},
{
"name": "placeholder",
"value": "Enter your last name"
}
]
}
],
"attributes": []
},
{
"type": "button",
"label": "Submit",
"children": [],
"attributes": [
{
"name": "type",
"value": "submit"
}
]
}
],
"attributes": [
{
"name": "method",
"value": "post"
},
{
"name": "action",
"value": "/submit-profile"
}
]
}콘텐츠 검토
콘텐츠 검토에서는 입력을 여러 범주로 분류하는 방식을 흔히 사용합니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26import OpenAI from "openai";
import { z } from "zod";
import { zodResponseFormat } from "openai/helpers/zod";
const openai = new OpenAI();
const ContentCompliance = z.object({
is_violating: z.boolean(),
category: z.enum(["violence", "sexual", "self_harm"]).nullable(),
explanation_if_violating: z.string().nullable(),
});
const completion = await openai.chat.completions.parse({
model: "gpt-6-astra",
messages: [
{
role: "system",
content:
"Determine if the user input violates specific guidelines and explain if they do.",
},
{ role: "user", content: "How do I prepare for a job interview?" },
],
response_format: zodResponseFormat(ContentCompliance, "content_compliance"),
});
const compliance = completion.choices[0].message.parsed;1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32from enum import Enum
from pydantic import BaseModel
from openai import OpenAI
client = OpenAI()
class Category(str, Enum):
violence = "violence"
sexual = "sexual"
self_harm = "self_harm"
class ContentCompliance(BaseModel):
is_violating: bool
category: Category | None
explanation_if_violating: str | None
completion = client.chat.completions.parse(
model="gpt-6-astra",
messages=[
{
"role": "system",
"content": "Determine if the user input violates specific guidelines and explain if they do.",
},
{"role": "user", "content": "How do I prepare for a job interview?"},
],
response_format=ContentCompliance,
)
compliance = completion.choices[0].message.parsed1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/shared"
)
func main() {
client := openai.NewClient()
schema := contentComplianceSchema()
completion, err := client.Chat.Completions.New(context.Background(), openai.ChatCompletionNewParams{
Model: "gpt-6-astra",
Messages: []openai.ChatCompletionMessageParamUnion{
openai.SystemMessage("Determine if the user input violates specific guidelines and explain if they do."),
openai.UserMessage("How do I prepare for a job interview?"),
},
ResponseFormat: openai.ChatCompletionNewParamsResponseFormatUnion{
OfJSONSchema: &shared.ResponseFormatJSONSchemaParam{JSONSchema: shared.ResponseFormatJSONSchemaJSONSchemaParam{
Name: "content_compliance", Description: openai.String("Determines if content is violating specific moderation rules"), Schema: schema, Strict: openai.Bool(true),
}},
},
})
if err != nil {
panic(err)
}
fmt.Println(completion.Choices[0].Message.Content)
}
func contentComplianceSchema() map[string]any {
return map[string]any{
"type": "object",
"properties": map[string]any{
"is_violating": map[string]any{"type": "boolean", "description": "Indicates if the content is violating guidelines"},
"category": map[string]any{"type": []string{"string", "null"}, "description": "Type of violation, if the content is violating guidelines. Null otherwise.", "enum": []any{"violence", "sexual", "self_harm", nil}},
"explanation_if_violating": map[string]any{"type": []string{"string", "null"}, "description": "Explanation of why the content is violating"},
},
"required": []string{"is_violating", "category", "explanation_if_violating"},
"additionalProperties": false,
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.chat.completions.ChatCompletionCreateParams;
import java.util.Arrays;
import java.util.List;
import java.util.Map;
Map<String, Object> schema =
Map.of(
"type",
"object",
"properties",
Map.of(
"is_violating",
Map.of(
"type", "boolean",
"description", "Whether the content violates the guidelines"),
"category",
Map.of(
"type", List.of("string", "null"),
"enum", Arrays.asList("violence", "sexual", "self_harm", null),
"description", "The violation category, or null when content is allowed"),
"explanation_if_violating",
Map.of(
"type",
List.of("string", "null"),
"description",
"Why the content violates the guidelines, or null")),
"required",
List.of("is_violating", "category", "explanation_if_violating"),
"additionalProperties",
false);
ChatCompletionCreateParams params =
ChatCompletionCreateParams.builder()
.model("gpt-6-astra")
.addSystemMessage(
"Determine whether the user input violates the guidelines and explain any violation.")
.addUserMessage("How do I prepare for a job interview?")
.putAdditionalBodyProperty(
"response_format",
JsonValue.from(
Map.of(
"type",
"json_schema",
"json_schema",
Map.of(
"name",
"content_compliance",
"description",
"Determines whether content violates moderation rules",
"strict",
true,
"schema",
schema))))
.build();
client.chat().completions().create(params).choices().stream()
.flatMap(choice -> choice.message().content().stream())
.forEach(System.out::println);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39using System.Text.Json;
using OpenAI.Chat;
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
string model = "gpt-6-astra";
ChatClient client = new(model, key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"is_violating": { "type": "boolean" },
"category": {
"type": ["string", "null"],
"enum": ["violence", "sexual", "self_harm", null]
},
"explanation_if_violating": { "type": ["string", "null"] }
},
"required": ["is_violating", "category", "explanation_if_violating"],
"additionalProperties": false
}
"""
);
ChatCompletionOptions options = new()
{
ResponseFormat = ChatResponseFormat.CreateJsonSchemaFormat(
"content_compliance",
schema,
jsonSchemaIsStrict: true
),
};
ChatCompletion completion = await client.CompleteChatAsync(
[new SystemChatMessage("Determine whether the user input violates content guidelines."), new UserChatMessage("How do I prepare for a job interview?")],
options
);
using JsonDocument parsed = JsonDocument.Parse(completion.Content[0].Text);
Console.WriteLine(parsed.RootElement);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48require "openai"
client = OpenAI::Client.new
compliance_schema = {
type: :object,
properties: {
is_violating: {
type: :boolean,
description: "Whether the content violates the guidelines"
},
category: {
type: %i[string null],
enum: ["violence", "sexual", "self_harm", nil],
description: "The violation category, or null when the content is allowed"
},
explanation_if_violating: {
type: %i[string null],
description: "Why the content violates the guidelines, or null"
}
},
required: %w[is_violating category explanation_if_violating],
additionalProperties: false
}
completion = client.chat.completions.create(
model: "gpt-6-astra",
messages: [
{
role: :system,
content: "Determine whether the user input violates the guidelines and explain any violation."
},
{
role: :user,
content: "How do I prepare for a job interview?"
}
],
response_format: {
type: :json_schema,
json_schema: {
name: "content_compliance",
description: "Determines whether content violates moderation rules",
strict: true,
schema: compliance_schema
}
}
)
puts(completion.choices.fetch(0).message.content)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44curl https://api.openai.com/v1/chat/completions \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6-astra",
"messages": [
{
"role": "system",
"content": "Determine if the user input violates specific guidelines and explain if they do."
},
{
"role": "user",
"content": "How do I prepare for a job interview?"
}
],
"response_format": {
"type": "json_schema",
"json_schema": {
"name": "content_compliance",
"description": "Determines if content is violating specific moderation rules",
"schema": {
"type": "object",
"properties": {
"is_violating": {
"type": "boolean",
"description": "Indicates if the content is violating guidelines"
},
"category": {
"type": ["string", "null"],
"description": "Type of violation, if the content is violating guidelines. Null otherwise.",
"enum": ["violence", "sexual", "self_harm"]
},
"explanation_if_violating": {
"type": ["string", "null"],
"description": "Explanation of why the content is violating"
}
},
"required": ["is_violating", "category", "explanation_if_violating"],
"additionalProperties": false
},
"strict": true
}
}
}'1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31import OpenAI from "openai";
import { zodTextFormat } from "openai/helpers/zod";
import { z } from "zod";
const openai = new OpenAI();
const ContentCompliance = z.object({
is_violating: z.boolean(),
category: z.enum(["violence", "sexual", "self_harm"]).nullable(),
explanation_if_violating: z.string().nullable(),
});
const response = await openai.responses.parse({
model: "gpt-6-astra",
input: [
{
role: "system",
content:
"Determine if the user input violates specific guidelines and explain if they do.",
},
{
role: "user",
content: "How do I prepare for a job interview?",
},
],
text: {
format: zodTextFormat(ContentCompliance, "content_compliance"),
},
});
const compliance = response.output_parsed;1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33from enum import Enum
from openai import OpenAI
from pydantic import BaseModel
client = OpenAI()
class Category(str, Enum):
violence = "violence"
sexual = "sexual"
self_harm = "self_harm"
class ContentCompliance(BaseModel):
is_violating: bool
category: Category | None
explanation_if_violating: str | None
response = client.responses.parse(
model="gpt-6-astra",
input=[
{
"role": "system",
"content": "Determine if the user input violates specific guidelines and explain if they do.",
},
{"role": "user", "content": "How do I prepare for a job interview?"},
],
text_format=ContentCompliance,
)
compliance = response.output_parsed1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/responses"
)
func main() {
client := openai.NewClient()
schema := contentComplianceSchema()
response, err := client.Responses.New(context.Background(), responses.ResponseNewParams{
Model: "gpt-6-astra",
Input: responses.ResponseNewParamsInputUnion{OfInputItemList: responses.ResponseInputParam{
responses.ResponseInputItemParamOfMessage("Determine if the user input violates specific guidelines and explain if they do.", responses.EasyInputMessageRoleSystem),
responses.ResponseInputItemParamOfMessage("How do I prepare for a job interview?", responses.EasyInputMessageRoleUser),
}},
Text: responses.ResponseTextConfigParam{Format: responses.ResponseFormatTextConfigUnionParam{
OfJSONSchema: &responses.ResponseFormatTextJSONSchemaConfigParam{
Name: "content_compliance", Description: openai.String("Determines if content is violating specific moderation rules"), Schema: schema, Strict: openai.Bool(true),
},
}},
})
if err != nil {
panic(err)
}
fmt.Println(response.OutputText())
}
func contentComplianceSchema() map[string]any {
return map[string]any{
"type": "object",
"properties": map[string]any{
"is_violating": map[string]any{"type": "boolean", "description": "Indicates if the content is violating guidelines"},
"category": map[string]any{"type": []string{"string", "null"}, "description": "Type of violation, if the content is violating guidelines. Null otherwise.", "enum": []any{"violence", "sexual", "self_harm", nil}},
"explanation_if_violating": map[string]any{"type": []string{"string", "null"}, "description": "Explanation of why the content is violating"},
},
"required": []string{"is_violating", "category", "explanation_if_violating"},
"additionalProperties": false,
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.responses.EasyInputMessage;
import com.openai.models.responses.ResponseCreateParams;
import com.openai.models.responses.ResponseFormatTextJsonSchemaConfig;
import com.openai.models.responses.ResponseInputItem;
import com.openai.models.responses.ResponseTextConfig;
import java.util.Arrays;
import java.util.List;
import java.util.Map;
Map<String, Object> schema =
Map.of(
"type",
"object",
"properties",
Map.of(
"is_violating",
Map.of(
"type", "boolean",
"description", "Whether the content violates the guidelines"),
"category",
Map.of(
"type", List.of("string", "null"),
"enum", Arrays.asList("violence", "sexual", "self_harm", null),
"description", "The violation category, or null when content is allowed"),
"explanation_if_violating",
Map.of(
"type",
List.of("string", "null"),
"description",
"Why the content violates the guidelines, or null")),
"required",
List.of("is_violating", "category", "explanation_if_violating"),
"additionalProperties",
false);
ResponseCreateParams params =
ResponseCreateParams.builder()
.model("gpt-6-astra")
.inputOfResponse(
List.of(
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.SYSTEM)
.content(
"Determine whether the user input violates the guidelines and explain any violation.")
.build()),
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.USER)
.content("How do I prepare for a job interview?")
.build())))
.text(
ResponseTextConfig.builder()
.format(
ResponseFormatTextJsonSchemaConfig.builder()
.name("content_compliance")
.description("Determines whether content violates moderation rules")
.strict(true)
.schema(
JsonValue.from(schema)
.convert(ResponseFormatTextJsonSchemaConfig.Schema.class))
.build())
.build())
.build();
client.responses().create(params).output().stream()
.flatMap(item -> item.message().stream())
.flatMap(message -> message.content().stream())
.flatMap(content -> content.outputText().stream())
.forEach(text -> System.out.println(text.text()));1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42using System.Text.Json;
using OpenAI.Responses;
#pragma warning disable OPENAI001
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ResponsesClient client = new(key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"is_violating": { "type": "boolean" },
"category": {
"type": ["string", "null"],
"enum": ["violence", "sexual", "self_harm", null]
},
"explanation_if_violating": { "type": ["string", "null"] }
},
"required": ["is_violating", "category", "explanation_if_violating"],
"additionalProperties": false
}
"""
);
CreateResponseOptions options = new()
{
Model = "gpt-6-astra",
TextOptions = new ResponseTextOptions
{
TextFormat = ResponseTextFormat.CreateJsonSchemaFormat(
"content_compliance",
schema,
jsonSchemaIsStrict: true
),
},
};
options.InputItems.Add(ResponseItem.CreateSystemMessageItem("Determine whether the user input violates content guidelines."));
options.InputItems.Add(ResponseItem.CreateUserMessageItem("How do I prepare for a job interview?"));
ResponseResult response = await client.CreateResponseAsync(options);
using JsonDocument parsed = JsonDocument.Parse(response.GetOutputText());
Console.WriteLine(parsed.RootElement);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48require "openai"
client = OpenAI::Client.new
compliance_schema = {
type: :object,
properties: {
is_violating: {
type: :boolean,
description: "Whether the content violates the guidelines"
},
category: {
type: %i[string null],
enum: ["violence", "sexual", "self_harm", nil],
description: "The violation category, or null when the content is allowed"
},
explanation_if_violating: {
type: %i[string null],
description: "Why the content violates the guidelines, or null"
}
},
required: %w[is_violating category explanation_if_violating],
additionalProperties: false
}
response = client.responses.create(
model: "gpt-6-astra",
input: [
{
role: :system,
content: "Determine whether the user input violates the guidelines and explain any violation."
},
{
role: :user,
content: "How do I prepare for a job interview?"
}
],
text: {
format: {
type: :json_schema,
name: "content_compliance",
description: "Determines whether content violates moderation rules",
strict: true,
schema: compliance_schema
}
}
)
puts(response.output_text)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44curl https://api.openai.com/v1/responses \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6-astra",
"input": [
{
"role": "system",
"content": "Determine if the user input violates specific guidelines and explain if they do."
},
{
"role": "user",
"content": "How do I prepare for a job interview?"
}
],
"text": {
"format": {
"type": "json_schema",
"name": "content_compliance",
"description": "Determines if content is violating specific moderation rules",
"schema": {
"type": "object",
"properties": {
"is_violating": {
"type": "boolean",
"description": "Indicates if the content is violating guidelines"
},
"category": {
"type": ["string", "null"],
"description": "Type of violation, if the content is violating guidelines. Null otherwise.",
"enum": ["violence", "sexual", "self_harm"]
},
"explanation_if_violating": {
"type": ["string", "null"],
"description": "Explanation of why the content is violating"
}
},
"required": ["is_violating", "category", "explanation_if_violating"],
"additionalProperties": false
},
"strict": true
}
}
}'응답 예시
{
"is_violating": false,
"category": null,
"explanation_if_violating": null
}구조화된 출력값 사용 방법: response_format
구조화된 출력값을 사용할 때는 새로운 SDK 헬퍼로 모델 출력을 원하는 형식으로 파싱하거나 JSON 스키마를 직접 지정할 수 있습니다.
참고: 파인튜닝된 모델에 특정 스키마로 처음 요청을 보내면 API가 스키마를 처리하는 데 추가 시간이 걸립니다. 이후 같은 스키마로 요청할 때는 추가 지연이 없습니다. 다른 모델에는 이 제한이 없습니다.
먼저 모델 응답이 따라야 할 JSON 스키마를 객체나 데이터 구조로 정의하세요. 이 가이드 상단의 예시를 참고하세요.
구조화된 출력값은 JSON 스키마의 기능을 대부분 지원하지만, 성능이나 기술상의 이유로 일부 기능은 사용할 수 없습니다. 자세한 내용은 지원되는 스키마를 참고하세요.
예를 들어 다음과 같이 객체를 정의할 수 있습니다.
1
2
3
4
5
6
7
8
9
10
11
12import { z } from "zod";
import { zodResponseFormat } from "openai/helpers/zod";
const Step = z.object({
explanation: z.string(),
output: z.string(),
});
const MathResponse = z.object({
steps: z.array(Step),
final_answer: z.string(),
});1
2
3
4
5
6
7
8
9
10
11from pydantic import BaseModel
class Step(BaseModel):
explanation: str
output: str
class MathResponse(BaseModel):
steps: list[Step]
final_answer: str데이터 구조 설계 팁
모델이 생성하는 응답의 품질을 최대한 높이려면 다음 방법을 권장합니다.
- 키 이름은 의미를 바로 알 수 있도록 명확하게 지으세요.
- 데이터 구조에서 중요한 키에는 명확한 제목과 설명을 작성하세요.
- 평가를 설계하고 실행해 사용 목적에 가장 적합한 데이터 구조를 찾으세요.
parse 메서드를 사용하면 JSON 응답을 앞에서 정의한 객체로 자동 파싱할 수 있습니다.
SDK가 내부적으로 데이터 구조에 해당하는 JSON 스키마를 API에 전달한 다음, 응답을 객체로 파싱합니다.
1
2
3
4
5
6
7
8
9
10
11
12const completion = await openai.chat.completions.parse({
model: "gpt-6-astra",
messages: [
{
role: "system",
content:
"You are a helpful math tutor. Guide the user through the solution step by step.",
},
{ role: "user", content: "how can I solve 8x + 7 = -23" },
],
response_format: zodResponseFormat(MathResponse, "math_response"),
});1
2
3
4
5
6
7
8
9
10
11completion = client.chat.completions.parse(
model="gpt-6-astra",
messages=[
{
"role": "system",
"content": "You are a helpful math tutor. Guide the user through the solution step by step.",
},
{"role": "user", "content": "how can I solve 8x + 7 = -23"},
],
response_format=MathResponse,
)경우에 따라 모델이 요청에 지정된 JSON 스키마에 맞는 유효한 응답을 생성하지 못할 수 있습니다.
예를 들어 모델이 안전상의 이유로 답변을 거부하거나, 최대 토큰 수에 도달해 응답이 중간에 끊길 때 이런 상황이 발생할 수 있습니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70try {
const completion = await openai.chat.completions.create({
model: "gpt-6-astra",
messages: [
{
role: "system",
content:
"You are a helpful math tutor. Guide the user through the solution step by step.",
},
{
role: "user",
content: "how can I solve 8x + 7 = -23",
},
],
store: true,
response_format: {
type: "json_schema",
json_schema: {
name: "math_response",
schema: {
type: "object",
properties: {
steps: {
type: "array",
items: {
type: "object",
properties: {
explanation: {
type: "string",
},
output: {
type: "string",
},
},
required: ["explanation", "output"],
additionalProperties: false,
},
},
final_answer: {
type: "string",
},
},
required: ["steps", "final_answer"],
additionalProperties: false,
},
strict: true,
},
},
max_completion_tokens: 1024,
});
if (completion.choices[0].finish_reason === "length") {
// Handle the case where the model did not return a complete response
throw new Error("Incomplete response");
}
const math_response = completion.choices[0].message;
if (math_response.refusal) {
// handle refusal
console.log(math_response.refusal);
} else if (math_response.content) {
console.log(math_response.content);
} else {
throw new Error("No response content");
}
} catch (e) {
// Handle edge cases
console.error(e);
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54try:
response = client.chat.completions.create(
model="gpt-6-astra",
messages=[
{
"role": "system",
"content": "You are a helpful math tutor. Guide the user through the solution step by step.",
},
{"role": "user", "content": "how can I solve 8x + 7 = -23"},
],
response_format={
"type": "json_schema",
"json_schema": {
"name": "math_response",
"strict": True,
"schema": {
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": {"type": "string"},
"output": {"type": "string"},
},
"required": ["explanation", "output"],
"additionalProperties": False,
},
},
"final_answer": {"type": "string"},
},
"required": ["steps", "final_answer"],
"additionalProperties": False,
},
},
},
max_completion_tokens=1024,
)
if response.choices[0].finish_reason == "length":
raise Exception("Incomplete response")
math_response = response.choices[0].message
if math_response.refusal:
print(math_response.refusal)
elif math_response.content:
print(math_response.content)
else:
raise Exception("No response content")
except Exception as e:
# handle errors like finish_reason, refusal, content_filter, etc.
print(e)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56package main
import (
"context"
"errors"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/shared"
)
func main() {
client := openai.NewClient()
completion, err := client.Chat.Completions.New(context.Background(), openai.ChatCompletionNewParams{
Model: "gpt-6-astra",
Messages: []openai.ChatCompletionMessageParamUnion{
openai.SystemMessage("You are a helpful math tutor. Guide the user through the solution step by step."),
openai.UserMessage("how can I solve 8x + 7 = -23"),
},
Store: openai.Bool(true),
MaxCompletionTokens: openai.Int(1024),
ResponseFormat: openai.ChatCompletionNewParamsResponseFormatUnion{
OfJSONSchema: &shared.ResponseFormatJSONSchemaParam{JSONSchema: shared.ResponseFormatJSONSchemaJSONSchemaParam{
Name: "math_response", Schema: mathSchema(), Strict: openai.Bool(true),
}},
},
})
if err != nil {
panic(err)
}
choice := completion.Choices[0]
if choice.FinishReason == "length" {
panic(errors.New("incomplete response"))
}
if choice.Message.Refusal != "" {
fmt.Println(choice.Message.Refusal)
return
}
if choice.Message.Content == "" {
panic(errors.New("no response content"))
}
fmt.Println(choice.Message.Content)
}
func mathSchema() map[string]any {
return map[string]any{
"type": "object",
"properties": map[string]any{
"steps": map[string]any{"type": "array", "items": map[string]any{"type": "object", "properties": map[string]any{"explanation": map[string]any{"type": "string"}, "output": map[string]any{"type": "string"}}, "required": []string{"explanation", "output"}, "additionalProperties": false}},
"final_answer": map[string]any{"type": "string"},
},
"required": []string{"steps", "final_answer"},
"additionalProperties": false,
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.chat.completions.ChatCompletion;
import com.openai.models.chat.completions.ChatCompletionCreateParams;
import java.util.List;
import java.util.Map;
Map<String, Object> stepSchema =
Map.of(
"type",
"object",
"properties",
Map.of(
"explanation", Map.of("type", "string"),
"output", Map.of("type", "string")),
"required",
List.of("explanation", "output"),
"additionalProperties",
false);
Map<String, Object> mathSchema =
Map.of(
"type",
"object",
"properties",
Map.of(
"steps", Map.of("type", "array", "items", stepSchema),
"final_answer", Map.of("type", "string")),
"required",
List.of("steps", "final_answer"),
"additionalProperties",
false);
ChatCompletionCreateParams params =
ChatCompletionCreateParams.builder()
.model("gpt-6-astra")
.addSystemMessage(
"You are a helpful math tutor. Guide the user through the solution step by step.")
.addUserMessage("How can I solve 8x + 7 = -23?")
.maxCompletionTokens(1024)
.store(true)
.putAdditionalBodyProperty(
"response_format",
JsonValue.from(
Map.of(
"type",
"json_schema",
"json_schema",
Map.of("name", "math_response", "strict", true, "schema", mathSchema))))
.build();
var choice = client.chat().completions().create(params).choices().get(0);
if (choice.finishReason().equals(ChatCompletion.Choice.FinishReason.LENGTH)) {
System.out.println("Incomplete response");
} else if (choice.message().refusal().isPresent()) {
System.out.println(choice.message().refusal().orElseThrow());
} else {
System.out.println(
choice
.message()
.content()
.orElseThrow(() -> new IllegalStateException("No response content")));
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64using OpenAI.Chat;
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
string model = "gpt-6-astra";
ChatClient client = new(model, key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": { "type": "string" },
"output": { "type": "string" }
},
"required": ["explanation", "output"],
"additionalProperties": false
}
},
"final_answer": { "type": "string" }
},
"required": ["steps", "final_answer"],
"additionalProperties": false
}
"""
);
ChatCompletionOptions options = new()
{
MaxOutputTokenCount = 300,
ResponseFormat = ChatResponseFormat.CreateJsonSchemaFormat(
"math_response",
schema,
jsonSchemaIsStrict: true
),
};
ChatCompletion completion = await client.CompleteChatAsync(
[new SystemChatMessage("You are a helpful math tutor. Guide the user through the solution step by step."), new UserChatMessage("How can I solve 8x + 7 = -23?")],
options
);
if (completion.FinishReason == ChatFinishReason.Length)
{
throw new InvalidOperationException("The structured response was incomplete.");
}
if (completion.FinishReason == ChatFinishReason.ContentFilter)
{
throw new InvalidOperationException("The structured response was interrupted by the content filter.");
}
if (!string.IsNullOrEmpty(completion.Refusal))
{
Console.WriteLine(completion.Refusal);
}
else if (completion.Content.Count > 0)
{
Console.WriteLine(completion.Content[0].Text);
}
else
{
throw new InvalidOperationException("The completion did not contain a response.");
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58require "openai"
client = OpenAI::Client.new
step_schema = {
type: :object,
properties: {
explanation: { type: :string },
output: { type: :string }
},
required: %w[explanation output],
additionalProperties: false
}
math_schema = {
type: :object,
properties: {
steps: {
type: :array,
items: step_schema
},
final_answer: { type: :string }
},
required: %w[steps final_answer],
additionalProperties: false
}
completion = client.chat.completions.create(
model: "gpt-6-astra",
messages: [
{
role: :system,
content: "You are a helpful math tutor. Guide the user through the solution step by step."
},
{
role: :user,
content: "How can I solve 8x + 7 = -23?"
}
],
max_completion_tokens: 1_024,
store: true,
response_format: {
type: :json_schema,
json_schema: {
name: "math_response",
strict: true,
schema: math_schema
}
}
)
choice = completion.choices.fetch(0)
if choice.finish_reason == OpenAI::Chat::ChatCompletion::Choice::FinishReason::LENGTH
raise "Incomplete response"
elsif choice.message.refusal
puts(choice.message.refusal)
else
content = choice.message.content or raise "No response content"
puts(content)
end구조화된 출력값을 사용하려면 다음과 같이 지정하면 됩니다.
response_format: { "type": "json_schema", "json_schema": … , "strict": true } text: { format: { type: "json_schema", "strict": true, "schema": … } } 사용 예시는 다음과 같습니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41const response = await openai.chat.completions.create({
model: "gpt-6-astra",
messages: [
{
role: "system",
content:
"You are a helpful math tutor. Guide the user through the solution step by step.",
},
{ role: "user", content: "how can I solve 8x + 7 = -23" },
],
store: true,
response_format: {
type: "json_schema",
json_schema: {
name: "math_response",
schema: {
type: "object",
properties: {
steps: {
type: "array",
items: {
type: "object",
properties: {
explanation: { type: "string" },
output: { type: "string" },
},
required: ["explanation", "output"],
additionalProperties: false,
},
},
final_answer: { type: "string" },
},
required: ["steps", "final_answer"],
additionalProperties: false,
},
strict: true,
},
},
});
console.log(response.choices[0].message.content);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39response = client.chat.completions.create(
model="gpt-6-astra",
messages=[
{
"role": "system",
"content": "You are a helpful math tutor. Guide the user through the solution step by step.",
},
{"role": "user", "content": "how can I solve 8x + 7 = -23"},
],
response_format={
"type": "json_schema",
"json_schema": {
"name": "math_response",
"schema": {
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": {"type": "string"},
"output": {"type": "string"},
},
"required": ["explanation", "output"],
"additionalProperties": False,
},
},
"final_answer": {"type": "string"},
},
"required": ["steps", "final_answer"],
"additionalProperties": False,
},
"strict": True,
},
},
)
print(response.choices[0].message.content)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/shared"
)
func main() {
client := openai.NewClient()
schema := mathSchema()
completion, err := client.Chat.Completions.New(context.Background(), openai.ChatCompletionNewParams{
Model: "gpt-6-astra",
Messages: []openai.ChatCompletionMessageParamUnion{
openai.SystemMessage("You are a helpful math tutor. Guide the user through the solution step by step."),
openai.UserMessage("how can I solve 8x + 7 = -23"),
},
Store: openai.Bool(true),
ResponseFormat: openai.ChatCompletionNewParamsResponseFormatUnion{
OfJSONSchema: &shared.ResponseFormatJSONSchemaParam{JSONSchema: shared.ResponseFormatJSONSchemaJSONSchemaParam{
Name: "math_response", Schema: schema, Strict: openai.Bool(true),
}},
},
})
if err != nil {
panic(err)
}
fmt.Println(completion.Choices[0].Message.Content)
}
func mathSchema() map[string]any {
return map[string]any{
"type": "object",
"properties": map[string]any{
"steps": map[string]any{"type": "array", "items": map[string]any{"type": "object", "properties": map[string]any{"explanation": map[string]any{"type": "string"}, "output": map[string]any{"type": "string"}}, "required": []string{"explanation", "output"}, "additionalProperties": false}},
"final_answer": map[string]any{"type": "string"},
},
"required": []string{"steps", "final_answer"},
"additionalProperties": false,
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.chat.completions.ChatCompletionCreateParams;
import java.util.List;
import java.util.Map;
Map<String, Object> stepSchema =
Map.of(
"type",
"object",
"properties",
Map.of(
"explanation", Map.of("type", "string"),
"output", Map.of("type", "string")),
"required",
List.of("explanation", "output"),
"additionalProperties",
false);
Map<String, Object> mathSchema =
Map.of(
"type",
"object",
"properties",
Map.of(
"steps", Map.of("type", "array", "items", stepSchema),
"final_answer", Map.of("type", "string")),
"required",
List.of("steps", "final_answer"),
"additionalProperties",
false);
ChatCompletionCreateParams params =
ChatCompletionCreateParams.builder()
.model("gpt-6-astra")
.addSystemMessage(
"You are a helpful math tutor. Guide the user through the solution step by step.")
.addUserMessage("How can I solve 8x + 7 = -23?")
.store(true)
.putAdditionalBodyProperty(
"response_format",
JsonValue.from(
Map.of(
"type",
"json_schema",
"json_schema",
Map.of("name", "math_response", "strict", true, "schema", mathSchema))))
.build();
client.chat().completions().create(params).choices().stream()
.flatMap(choice -> choice.message().content().stream())
.forEach(System.out::println);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46using System.Text.Json;
using OpenAI.Chat;
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
string model = "gpt-6-astra";
ChatClient client = new(model, key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": { "type": "string" },
"output": { "type": "string" }
},
"required": ["explanation", "output"],
"additionalProperties": false
}
},
"final_answer": { "type": "string" }
},
"required": ["steps", "final_answer"],
"additionalProperties": false
}
"""
);
ChatCompletionOptions options = new()
{
ResponseFormat = ChatResponseFormat.CreateJsonSchemaFormat(
"math_response",
schema,
jsonSchemaIsStrict: true
),
};
ChatCompletion completion = await client.CompleteChatAsync(
[new SystemChatMessage("You are a helpful math tutor. Guide the user through the solution step by step."), new UserChatMessage("How can I solve 8x + 7 = -23?")],
options
);
using JsonDocument parsed = JsonDocument.Parse(completion.Content[0].Text);
Console.WriteLine(parsed.RootElement);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48require "openai"
client = OpenAI::Client.new
math_schema = {
type: :object,
properties: {
steps: {
type: :array,
items: {
type: :object,
properties: {
explanation: { type: :string },
output: { type: :string }
},
required: %w[explanation output],
additionalProperties: false
}
},
final_answer: { type: :string }
},
required: %w[steps final_answer],
additionalProperties: false
}
completion = client.chat.completions.create(
model: "gpt-6-astra",
messages: [
{
role: :system,
content: "You are a helpful math tutor. Guide the user through the solution step by step."
},
{
role: :user,
content: "How can I solve 8x + 7 = -23?"
}
],
store: true,
response_format: {
type: :json_schema,
json_schema: {
name: "math_response",
strict: true,
schema: math_schema
}
}
)
puts(completion.choices.fetch(0).message.content)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43curl https://api.openai.com/v1/chat/completions \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6-astra",
"messages": [
{
"role": "system",
"content": "You are a helpful math tutor. Guide the user through the solution step by step."
},
{
"role": "user",
"content": "how can I solve 8x + 7 = -23"
}
],
"response_format": {
"type": "json_schema",
"json_schema": {
"name": "math_response",
"schema": {
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": { "type": "string" },
"output": { "type": "string" }
},
"required": ["explanation", "output"],
"additionalProperties": false
}
},
"final_answer": { "type": "string" }
},
"required": ["steps", "final_answer"],
"additionalProperties": false
},
"strict": true
}
}
}'1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40const response = await openai.responses.create({
model: "gpt-6-astra",
input: [
{
role: "system",
content:
"You are a helpful math tutor. Guide the user through the solution step by step.",
},
{ role: "user", content: "how can I solve 8x + 7 = -23" },
],
text: {
format: {
type: "json_schema",
name: "math_response",
schema: {
type: "object",
properties: {
steps: {
type: "array",
items: {
type: "object",
properties: {
explanation: { type: "string" },
output: { type: "string" },
},
required: ["explanation", "output"],
additionalProperties: false,
},
},
final_answer: { type: "string" },
},
required: ["steps", "final_answer"],
additionalProperties: false,
},
strict: true,
},
},
});
console.log(response.output_text);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39response = client.responses.create(
model="gpt-6-astra",
input=[
{
"role": "system",
"content": "You are a helpful math tutor. Guide the user through the solution step by step.",
},
{"role": "user", "content": "how can I solve 8x + 7 = -23"},
],
text={
"format": {
"type": "json_schema",
"name": "math_response",
"schema": {
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": {"type": "string"},
"output": {"type": "string"},
},
"required": ["explanation", "output"],
"additionalProperties": False,
},
},
"final_answer": {"type": "string"},
},
"required": ["steps", "final_answer"],
"additionalProperties": False,
},
"strict": True,
},
},
)
print(response.output_text)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/responses"
)
func main() {
client := openai.NewClient()
response, err := client.Responses.New(context.Background(), responses.ResponseNewParams{
Model: "gpt-6-astra",
Input: responses.ResponseNewParamsInputUnion{OfInputItemList: responses.ResponseInputParam{
responses.ResponseInputItemParamOfMessage(
responses.ResponseInputMessageContentListParam{responses.ResponseInputContentParamOfInputText("You are a helpful math tutor. Guide the user through the solution step by step.")},
responses.EasyInputMessageRoleSystem,
),
responses.ResponseInputItemParamOfMessage(
responses.ResponseInputMessageContentListParam{responses.ResponseInputContentParamOfInputText("how can I solve 8x + 7 = -23")},
responses.EasyInputMessageRoleUser,
),
}},
Text: responses.ResponseTextConfigParam{Format: responses.ResponseFormatTextConfigUnionParam{
OfJSONSchema: &responses.ResponseFormatTextJSONSchemaConfigParam{Name: "math_response", Schema: mathSchema(), Strict: openai.Bool(true)},
}},
})
if err != nil {
panic(err)
}
fmt.Println(response.OutputText())
}
func mathSchema() map[string]any {
return map[string]any{
"type": "object",
"properties": map[string]any{
"steps": map[string]any{"type": "array", "items": map[string]any{"type": "object", "properties": map[string]any{"explanation": map[string]any{"type": "string"}, "output": map[string]any{"type": "string"}}, "required": []string{"explanation", "output"}, "additionalProperties": false}},
"final_answer": map[string]any{"type": "string"},
},
"required": []string{"steps", "final_answer"},
"additionalProperties": false,
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.responses.EasyInputMessage;
import com.openai.models.responses.ResponseCreateParams;
import com.openai.models.responses.ResponseFormatTextJsonSchemaConfig;
import com.openai.models.responses.ResponseInputItem;
import com.openai.models.responses.ResponseTextConfig;
import java.util.List;
import java.util.Map;
Map<String, Object> schema =
Map.of(
"type",
"object",
"properties",
Map.of(
"steps",
Map.of(
"type",
"array",
"items",
Map.of(
"type",
"object",
"properties",
Map.of(
"explanation", Map.of("type", "string"),
"output", Map.of("type", "string")),
"required",
List.of("explanation", "output"),
"additionalProperties",
false)),
"final_answer", Map.of("type", "string")),
"required",
List.of("steps", "final_answer"),
"additionalProperties",
false);
ResponseCreateParams params =
ResponseCreateParams.builder()
.model("gpt-6-astra")
.inputOfResponse(
List.of(
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.SYSTEM)
.content(
"You are a helpful math tutor. Guide the user through the solution step by step.")
.build()),
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.USER)
.content("How can I solve 8x + 7 = -23?")
.build())))
.text(
ResponseTextConfig.builder()
.format(
ResponseFormatTextJsonSchemaConfig.builder()
.name("math_response")
.strict(true)
.schema(
JsonValue.from(schema)
.convert(ResponseFormatTextJsonSchemaConfig.Schema.class))
.build())
.build())
.build();
client.responses().create(params).output().stream()
.flatMap(item -> item.message().stream())
.flatMap(message -> message.content().stream())
.flatMap(content -> content.outputText().stream())
.forEach(text -> System.out.println(text.text()));1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49using System.Text.Json;
using OpenAI.Responses;
#pragma warning disable OPENAI001
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ResponsesClient client = new(key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": { "type": "string" },
"output": { "type": "string" }
},
"required": ["explanation", "output"],
"additionalProperties": false
}
},
"final_answer": { "type": "string" }
},
"required": ["steps", "final_answer"],
"additionalProperties": false
}
"""
);
CreateResponseOptions options = new()
{
Model = "gpt-6-astra",
TextOptions = new ResponseTextOptions
{
TextFormat = ResponseTextFormat.CreateJsonSchemaFormat(
"math_response",
schema,
jsonSchemaIsStrict: true
),
},
};
options.InputItems.Add(ResponseItem.CreateSystemMessageItem("You are a helpful math tutor. Guide the user through the solution step by step."));
options.InputItems.Add(ResponseItem.CreateUserMessageItem("How can I solve 8x + 7 = -23?"));
ResponseResult response = await client.CreateResponseAsync(options);
using JsonDocument parsed = JsonDocument.Parse(response.GetOutputText());
Console.WriteLine(parsed.RootElement);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47require "openai"
client = OpenAI::Client.new
math_schema = {
type: :object,
properties: {
steps: {
type: :array,
items: {
type: :object,
properties: {
explanation: { type: :string },
output: { type: :string }
},
required: %w[explanation output],
additionalProperties: false
}
},
final_answer: { type: :string }
},
required: %w[steps final_answer],
additionalProperties: false
}
response = client.responses.create(
model: "gpt-6-astra",
input: [
{
role: :system,
content: "You are a helpful math tutor. Guide the user through the solution step by step."
},
{
role: :user,
content: "How can I solve 8x + 7 = -23?"
}
],
text: {
format: {
type: :json_schema,
name: "math_response",
strict: true,
schema: math_schema
}
}
)
puts(response.output_text)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43curl https://api.openai.com/v1/responses \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6-astra",
"input": [
{
"role": "system",
"content": "You are a helpful math tutor. Guide the user through the solution step by step."
},
{
"role": "user",
"content": "how can I solve 8x + 7 = -23"
}
],
"text": {
"format": {
"type": "json_schema",
"name": "math_response",
"schema": {
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": { "type": "string" },
"output": { "type": "string" }
},
"required": ["explanation", "output"],
"additionalProperties": false
}
},
"final_answer": { "type": "string" }
},
"required": ["steps", "final_answer"],
"additionalProperties": false
},
"strict": true
}
}
}'참고: 각 스키마를 처음 사용하는 요청에서는 API가 해당 스키마를 처리하는 데 시간이 더 걸립니다. 이후 같은 스키마를 사용하는 요청에서는 이로 인한 추가 지연이 없습니다.
모델이 지정된 JSON 스키마에 맞는 유효한 응답을 생성하지 못할 때가 있습니다.
예를 들어 모델이 안전상의 이유로 답변을 거부하거나 토큰 수가 한도에 도달해 응답이 중간에 끊기는 경우입니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70try {
const completion = await openai.chat.completions.create({
model: "gpt-6-astra",
messages: [
{
role: "system",
content:
"You are a helpful math tutor. Guide the user through the solution step by step.",
},
{
role: "user",
content: "how can I solve 8x + 7 = -23",
},
],
store: true,
response_format: {
type: "json_schema",
json_schema: {
name: "math_response",
schema: {
type: "object",
properties: {
steps: {
type: "array",
items: {
type: "object",
properties: {
explanation: {
type: "string",
},
output: {
type: "string",
},
},
required: ["explanation", "output"],
additionalProperties: false,
},
},
final_answer: {
type: "string",
},
},
required: ["steps", "final_answer"],
additionalProperties: false,
},
strict: true,
},
},
max_completion_tokens: 1024,
});
if (completion.choices[0].finish_reason === "length") {
// Handle the case where the model did not return a complete response
throw new Error("Incomplete response");
}
const math_response = completion.choices[0].message;
if (math_response.refusal) {
// handle refusal
console.log(math_response.refusal);
} else if (math_response.content) {
console.log(math_response.content);
} else {
throw new Error("No response content");
}
} catch (e) {
// Handle edge cases
console.error(e);
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54try:
response = client.chat.completions.create(
model="gpt-6-astra",
messages=[
{
"role": "system",
"content": "You are a helpful math tutor. Guide the user through the solution step by step.",
},
{"role": "user", "content": "how can I solve 8x + 7 = -23"},
],
response_format={
"type": "json_schema",
"json_schema": {
"name": "math_response",
"strict": True,
"schema": {
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": {"type": "string"},
"output": {"type": "string"},
},
"required": ["explanation", "output"],
"additionalProperties": False,
},
},
"final_answer": {"type": "string"},
},
"required": ["steps", "final_answer"],
"additionalProperties": False,
},
},
},
max_completion_tokens=1024,
)
if response.choices[0].finish_reason == "length":
raise Exception("Incomplete response")
math_response = response.choices[0].message
if math_response.refusal:
print(math_response.refusal)
elif math_response.content:
print(math_response.content)
else:
raise Exception("No response content")
except Exception as e:
# handle errors like finish_reason, refusal, content_filter, etc.
print(e)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56package main
import (
"context"
"errors"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/shared"
)
func main() {
client := openai.NewClient()
completion, err := client.Chat.Completions.New(context.Background(), openai.ChatCompletionNewParams{
Model: "gpt-6-astra",
Messages: []openai.ChatCompletionMessageParamUnion{
openai.SystemMessage("You are a helpful math tutor. Guide the user through the solution step by step."),
openai.UserMessage("how can I solve 8x + 7 = -23"),
},
Store: openai.Bool(true),
MaxCompletionTokens: openai.Int(1024),
ResponseFormat: openai.ChatCompletionNewParamsResponseFormatUnion{
OfJSONSchema: &shared.ResponseFormatJSONSchemaParam{JSONSchema: shared.ResponseFormatJSONSchemaJSONSchemaParam{
Name: "math_response", Schema: mathSchema(), Strict: openai.Bool(true),
}},
},
})
if err != nil {
panic(err)
}
choice := completion.Choices[0]
if choice.FinishReason == "length" {
panic(errors.New("incomplete response"))
}
if choice.Message.Refusal != "" {
fmt.Println(choice.Message.Refusal)
return
}
if choice.Message.Content == "" {
panic(errors.New("no response content"))
}
fmt.Println(choice.Message.Content)
}
func mathSchema() map[string]any {
return map[string]any{
"type": "object",
"properties": map[string]any{
"steps": map[string]any{"type": "array", "items": map[string]any{"type": "object", "properties": map[string]any{"explanation": map[string]any{"type": "string"}, "output": map[string]any{"type": "string"}}, "required": []string{"explanation", "output"}, "additionalProperties": false}},
"final_answer": map[string]any{"type": "string"},
},
"required": []string{"steps", "final_answer"},
"additionalProperties": false,
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.chat.completions.ChatCompletion;
import com.openai.models.chat.completions.ChatCompletionCreateParams;
import java.util.List;
import java.util.Map;
Map<String, Object> stepSchema =
Map.of(
"type",
"object",
"properties",
Map.of(
"explanation", Map.of("type", "string"),
"output", Map.of("type", "string")),
"required",
List.of("explanation", "output"),
"additionalProperties",
false);
Map<String, Object> mathSchema =
Map.of(
"type",
"object",
"properties",
Map.of(
"steps", Map.of("type", "array", "items", stepSchema),
"final_answer", Map.of("type", "string")),
"required",
List.of("steps", "final_answer"),
"additionalProperties",
false);
ChatCompletionCreateParams params =
ChatCompletionCreateParams.builder()
.model("gpt-6-astra")
.addSystemMessage(
"You are a helpful math tutor. Guide the user through the solution step by step.")
.addUserMessage("How can I solve 8x + 7 = -23?")
.maxCompletionTokens(1024)
.store(true)
.putAdditionalBodyProperty(
"response_format",
JsonValue.from(
Map.of(
"type",
"json_schema",
"json_schema",
Map.of("name", "math_response", "strict", true, "schema", mathSchema))))
.build();
var choice = client.chat().completions().create(params).choices().get(0);
if (choice.finishReason().equals(ChatCompletion.Choice.FinishReason.LENGTH)) {
System.out.println("Incomplete response");
} else if (choice.message().refusal().isPresent()) {
System.out.println(choice.message().refusal().orElseThrow());
} else {
System.out.println(
choice
.message()
.content()
.orElseThrow(() -> new IllegalStateException("No response content")));
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64using OpenAI.Chat;
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
string model = "gpt-6-astra";
ChatClient client = new(model, key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": { "type": "string" },
"output": { "type": "string" }
},
"required": ["explanation", "output"],
"additionalProperties": false
}
},
"final_answer": { "type": "string" }
},
"required": ["steps", "final_answer"],
"additionalProperties": false
}
"""
);
ChatCompletionOptions options = new()
{
MaxOutputTokenCount = 300,
ResponseFormat = ChatResponseFormat.CreateJsonSchemaFormat(
"math_response",
schema,
jsonSchemaIsStrict: true
),
};
ChatCompletion completion = await client.CompleteChatAsync(
[new SystemChatMessage("You are a helpful math tutor. Guide the user through the solution step by step."), new UserChatMessage("How can I solve 8x + 7 = -23?")],
options
);
if (completion.FinishReason == ChatFinishReason.Length)
{
throw new InvalidOperationException("The structured response was incomplete.");
}
if (completion.FinishReason == ChatFinishReason.ContentFilter)
{
throw new InvalidOperationException("The structured response was interrupted by the content filter.");
}
if (!string.IsNullOrEmpty(completion.Refusal))
{
Console.WriteLine(completion.Refusal);
}
else if (completion.Content.Count > 0)
{
Console.WriteLine(completion.Content[0].Text);
}
else
{
throw new InvalidOperationException("The completion did not contain a response.");
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58require "openai"
client = OpenAI::Client.new
step_schema = {
type: :object,
properties: {
explanation: { type: :string },
output: { type: :string }
},
required: %w[explanation output],
additionalProperties: false
}
math_schema = {
type: :object,
properties: {
steps: {
type: :array,
items: step_schema
},
final_answer: { type: :string }
},
required: %w[steps final_answer],
additionalProperties: false
}
completion = client.chat.completions.create(
model: "gpt-6-astra",
messages: [
{
role: :system,
content: "You are a helpful math tutor. Guide the user through the solution step by step."
},
{
role: :user,
content: "How can I solve 8x + 7 = -23?"
}
],
max_completion_tokens: 1_024,
store: true,
response_format: {
type: :json_schema,
json_schema: {
name: "math_response",
strict: true,
schema: math_schema
}
}
)
choice = completion.choices.fetch(0)
if choice.finish_reason == OpenAI::Chat::ChatCompletion::Choice::FinishReason::LENGTH
raise "Incomplete response"
elsif choice.message.refusal
puts(choice.message.refusal)
else
content = choice.message.content or raise "No response content"
puts(content)
end1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77try {
const response = await openai.responses.create({
model: "gpt-6-astra",
input: [
{
role: "system",
content:
"You are a helpful math tutor. Guide the user through the solution step by step.",
},
{
role: "user",
content: "how can I solve 8x + 7 = -23",
},
],
max_output_tokens: 1024,
text: {
format: {
type: "json_schema",
name: "math_response",
schema: {
type: "object",
properties: {
steps: {
type: "array",
items: {
type: "object",
properties: {
explanation: {
type: "string",
},
output: {
type: "string",
},
},
required: ["explanation", "output"],
additionalProperties: false,
},
},
final_answer: {
type: "string",
},
},
required: ["steps", "final_answer"],
additionalProperties: false,
},
strict: true,
},
},
});
if (
response.status === "incomplete" &&
response.incomplete_details.reason === "max_output_tokens"
) {
// Handle the case where the model did not return a complete response
throw new Error("Incomplete response");
}
const message = response.output.find((item) => item.type === "message");
const math_response = message?.content[0];
if (!math_response) {
throw new Error("No response content");
}
if (math_response.type === "refusal") {
// handle refusal
console.log(math_response.refusal);
} else if (math_response.type === "output_text") {
console.log(math_response.text);
} else {
throw new Error("No response content");
}
} catch (e) {
// Handle edge cases
console.error(e);
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61try:
response = client.responses.create(
model="gpt-6-astra",
input=[
{
"role": "system",
"content": "You are a helpful math tutor. Guide the user through the solution step by step.",
},
{"role": "user", "content": "how can I solve 8x + 7 = -23"},
],
text={
"format": {
"type": "json_schema",
"name": "math_response",
"strict": True,
"schema": {
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": {"type": "string"},
"output": {"type": "string"},
},
"required": ["explanation", "output"],
"additionalProperties": False,
},
},
"final_answer": {"type": "string"},
},
"required": ["steps", "final_answer"],
"additionalProperties": False,
},
},
},
max_output_tokens=1024,
)
if (
response.status == "incomplete"
and response.incomplete_details.reason == "max_output_tokens"
):
raise Exception("Incomplete response")
message = next((item for item in response.output if item.type == "message"), None)
math_response = message.content[0] if message and message.content else None
if not math_response:
raise Exception("No response content")
if math_response.type == "refusal":
print(math_response.refusal)
elif math_response.type == "output_text":
print(math_response.text)
else:
raise Exception("No response content")
except Exception as e:
# handle errors like finish_reason, refusal, content_filter, etc.
print(e)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66package main
import (
"context"
"errors"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/responses"
)
func main() {
client := openai.NewClient()
response, err := client.Responses.New(context.Background(), responses.ResponseNewParams{
Model: "gpt-6-astra",
Input: responses.ResponseNewParamsInputUnion{OfInputItemList: responses.ResponseInputParam{
responses.ResponseInputItemParamOfMessage(
responses.ResponseInputMessageContentListParam{responses.ResponseInputContentParamOfInputText("You are a helpful math tutor. Guide the user through the solution step by step.")},
responses.EasyInputMessageRoleSystem,
),
responses.ResponseInputItemParamOfMessage(
responses.ResponseInputMessageContentListParam{responses.ResponseInputContentParamOfInputText("how can I solve 8x + 7 = -23")},
responses.EasyInputMessageRoleUser,
),
}},
MaxOutputTokens: openai.Int(1024),
Text: responses.ResponseTextConfigParam{Format: responses.ResponseFormatTextConfigUnionParam{
OfJSONSchema: &responses.ResponseFormatTextJSONSchemaConfigParam{Name: "math_response", Schema: mathSchema(), Strict: openai.Bool(true)},
}},
})
if err != nil {
panic(err)
}
if response.Status == "incomplete" {
panic(errors.New("incomplete response"))
}
for _, output := range response.Output {
if output.Type != "message" {
continue
}
for _, content := range output.AsMessage().Content {
if content.Type == "refusal" {
fmt.Println(content.AsRefusal().Refusal)
return
}
if content.Type == "output_text" {
fmt.Println(content.AsOutputText().Text)
return
}
}
}
panic(errors.New("no response content"))
}
func mathSchema() map[string]any {
return map[string]any{
"type": "object",
"properties": map[string]any{
"steps": map[string]any{"type": "array", "items": map[string]any{"type": "object", "properties": map[string]any{"explanation": map[string]any{"type": "string"}, "output": map[string]any{"type": "string"}}, "required": []string{"explanation", "output"}, "additionalProperties": false}},
"final_answer": map[string]any{"type": "string"},
},
"required": []string{"steps", "final_answer"},
"additionalProperties": false,
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.responses.EasyInputMessage;
import com.openai.models.responses.ResponseCreateParams;
import com.openai.models.responses.ResponseFormatTextJsonSchemaConfig;
import com.openai.models.responses.ResponseInputItem;
import com.openai.models.responses.ResponseStatus;
import com.openai.models.responses.ResponseTextConfig;
import java.util.List;
import java.util.Map;
ResponseCreateParams params =
ResponseCreateParams.builder()
.model("gpt-6-astra")
.inputOfResponse(
List.of(
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.SYSTEM)
.content(
"You are a helpful math tutor. Guide the user through the solution step by step.")
.build()),
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.USER)
.content("How can I solve 8x + 7 = -23?")
.build())))
.text(
ResponseTextConfig.builder()
.format(
ResponseFormatTextJsonSchemaConfig.builder()
.name("math_response")
.strict(true)
.schema(
ResponseFormatTextJsonSchemaConfig.Schema.builder()
.putAdditionalProperty("type", JsonValue.from("object"))
.putAdditionalProperty(
"properties",
JsonValue.from(
Map.of(
"steps",
Map.of(
"type",
"array",
"items",
Map.of(
"type",
"object",
"properties",
Map.of(
"explanation",
Map.of("type", "string"),
"output",
Map.of("type", "string")),
"required",
List.of("explanation", "output"),
"additionalProperties",
false)),
"final_answer",
Map.of("type", "string"))))
.putAdditionalProperty(
"required",
JsonValue.from(List.of("steps", "final_answer")))
.putAdditionalProperty(
"additionalProperties", JsonValue.from(false))
.build())
.build())
.build())
.maxOutputTokens(1_024L)
.build();
var response = client.responses().create(params);
if (response.status().filter(ResponseStatus.INCOMPLETE::equals).isPresent()) {
throw new IllegalStateException("Incomplete response");
}
var content =
response.output().stream()
.flatMap(item -> item.message().stream())
.flatMap(message -> message.content().stream())
.findFirst()
.orElseThrow(() -> new IllegalStateException("No response content"));
if (content.refusal().isPresent()) {
System.out.println(content.refusal().orElseThrow().refusal());
} else {
System.out.println(
content
.outputText()
.orElseThrow(() -> new IllegalStateException("No response content"))
.text());
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68using OpenAI.Responses;
#pragma warning disable OPENAI001
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ResponsesClient client = new(key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": { "type": "string" },
"output": { "type": "string" }
},
"required": ["explanation", "output"],
"additionalProperties": false
}
},
"final_answer": { "type": "string" }
},
"required": ["steps", "final_answer"],
"additionalProperties": false
}
"""
);
CreateResponseOptions options = new()
{
Model = "gpt-6-astra",
MaxOutputTokenCount = 300,
TextOptions = new ResponseTextOptions
{
TextFormat = ResponseTextFormat.CreateJsonSchemaFormat(
"math_response",
schema,
jsonSchemaIsStrict: true
),
},
};
options.InputItems.Add(ResponseItem.CreateSystemMessageItem("You are a helpful math tutor. Guide the user through the solution step by step."));
options.InputItems.Add(ResponseItem.CreateUserMessageItem("How can I solve 8x + 7 = -23?"));
ResponseResult response = await client.CreateResponseAsync(options);
if (
response.Status == ResponseStatus.Incomplete
&& response.IncompleteStatusDetails?.Reason == ResponseIncompleteStatusReason.MaxOutputTokens
)
{
throw new InvalidOperationException("The structured response was incomplete.");
}
if (
response.Status == ResponseStatus.Incomplete
&& response.IncompleteStatusDetails?.Reason == ResponseIncompleteStatusReason.ContentFilter
)
{
throw new InvalidOperationException("The structured response was interrupted by the content filter.");
}
MessageResponseItem message = response.OutputItems.OfType<MessageResponseItem>().FirstOrDefault()
?? throw new InvalidOperationException("The response did not include an output message.");
ResponseContentPart content = message.Content.FirstOrDefault()
?? throw new InvalidOperationException("The response did not include output content.");
Console.WriteLine(
content.Kind == ResponseContentPartKind.Refusal ? content.Refusal : content.Text
);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65require "openai"
client = OpenAI::Client.new
step_schema = {
type: :object,
properties: {
explanation: { type: :string },
output: { type: :string }
},
required: %w[explanation output],
additionalProperties: false
}
math_schema = {
type: :object,
properties: {
steps: {
type: :array,
items: step_schema
},
final_answer: { type: :string }
},
required: %w[steps final_answer],
additionalProperties: false
}
response = client.responses.create(
model: "gpt-6-astra",
input: [
{
role: :system,
content: "You are a helpful math tutor. Guide the user through the solution step by step."
},
{
role: :user,
content: "How can I solve 8x + 7 = -23?"
}
],
max_output_tokens: 1_024,
text: {
format: {
type: :json_schema,
name: "math_response",
strict: true,
schema: math_schema
}
}
)
if response.status == OpenAI::Responses::ResponseStatus::INCOMPLETE
raise "Incomplete response"
end
message = response.output.find do |item|
item.is_a?(OpenAI::Models::Responses::ResponseOutputMessage)
end
unless message.is_a?(OpenAI::Models::Responses::ResponseOutputMessage)
raise "No response message"
end
content = message.content.fetch(0)
if content.is_a?(OpenAI::Models::Responses::ResponseOutputRefusal)
puts(content.refusal)
else
puts(content.text)
end응답에 스키마와 일치하는 JSON이 포함되어 있는지 확인한 뒤, 해당 JSON을 파싱해 사용 중인 언어의 기본 데이터 구조로 변환하세요. 타입을 지원하는 언어에서는 데이터 구조에 맞는 타입이나 클래스로 데이터를 모델링할 수도 있습니다.
다음은 예시입니다.
1
2
3
4
5// The request that produces `response` appears earlier in this guide.
const content = response.choices[0].message.content;
if (!content) throw new Error("The response did not contain JSON output.");
const solution = JSON.parse(content);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18from pydantic import BaseModel, ValidationError
class Step(BaseModel):
explanation: str
output: str
class Solution(BaseModel):
steps: list[Step]
final_answer: str
try:
solution = Solution.model_validate_json(response.choices[0].message.content)
print(solution)
except ValidationError as error:
print(error.json())1
2
3
4
5
6
7
8
9
10
11var solution struct {
Steps []struct {
Explanation string `json:"explanation"`
Output string `json:"output"`
} `json:"steps"`
FinalAnswer string `json:"final_answer"`
}
if err := json.Unmarshal([]byte(completion.Choices[0].Message.Content), &solution); err != nil {
return err
}
fmt.Println(solution)1System.out.println(new ObjectMapper().readTree(content));1
2
3
4
5using System.Text.Json;
using OpenAI.Chat;
using JsonDocument solution = JsonDocument.Parse(completion.Content[0].Text);
Console.WriteLine(solution.RootElement);1
2
3
4
5content = completion.choices.fetch(0).message.content
raise "Response did not contain JSON output." if content.nil?
solution = JSON.parse(content)
puts(solution)구조화된 출력값 사용 방법: text.format
구조화된 출력값을 사용하려면 다음과 같이 지정하면 됩니다.
response_format: { "type": "json_schema", "json_schema": … , "strict": true } text: { format: { type: "json_schema", "strict": true, "schema": … } } 사용 예시는 다음과 같습니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41const response = await openai.chat.completions.create({
model: "gpt-6-astra",
messages: [
{
role: "system",
content:
"You are a helpful math tutor. Guide the user through the solution step by step.",
},
{ role: "user", content: "how can I solve 8x + 7 = -23" },
],
store: true,
response_format: {
type: "json_schema",
json_schema: {
name: "math_response",
schema: {
type: "object",
properties: {
steps: {
type: "array",
items: {
type: "object",
properties: {
explanation: { type: "string" },
output: { type: "string" },
},
required: ["explanation", "output"],
additionalProperties: false,
},
},
final_answer: { type: "string" },
},
required: ["steps", "final_answer"],
additionalProperties: false,
},
strict: true,
},
},
});
console.log(response.choices[0].message.content);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39response = client.chat.completions.create(
model="gpt-6-astra",
messages=[
{
"role": "system",
"content": "You are a helpful math tutor. Guide the user through the solution step by step.",
},
{"role": "user", "content": "how can I solve 8x + 7 = -23"},
],
response_format={
"type": "json_schema",
"json_schema": {
"name": "math_response",
"schema": {
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": {"type": "string"},
"output": {"type": "string"},
},
"required": ["explanation", "output"],
"additionalProperties": False,
},
},
"final_answer": {"type": "string"},
},
"required": ["steps", "final_answer"],
"additionalProperties": False,
},
"strict": True,
},
},
)
print(response.choices[0].message.content)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/shared"
)
func main() {
client := openai.NewClient()
schema := mathSchema()
completion, err := client.Chat.Completions.New(context.Background(), openai.ChatCompletionNewParams{
Model: "gpt-6-astra",
Messages: []openai.ChatCompletionMessageParamUnion{
openai.SystemMessage("You are a helpful math tutor. Guide the user through the solution step by step."),
openai.UserMessage("how can I solve 8x + 7 = -23"),
},
Store: openai.Bool(true),
ResponseFormat: openai.ChatCompletionNewParamsResponseFormatUnion{
OfJSONSchema: &shared.ResponseFormatJSONSchemaParam{JSONSchema: shared.ResponseFormatJSONSchemaJSONSchemaParam{
Name: "math_response", Schema: schema, Strict: openai.Bool(true),
}},
},
})
if err != nil {
panic(err)
}
fmt.Println(completion.Choices[0].Message.Content)
}
func mathSchema() map[string]any {
return map[string]any{
"type": "object",
"properties": map[string]any{
"steps": map[string]any{"type": "array", "items": map[string]any{"type": "object", "properties": map[string]any{"explanation": map[string]any{"type": "string"}, "output": map[string]any{"type": "string"}}, "required": []string{"explanation", "output"}, "additionalProperties": false}},
"final_answer": map[string]any{"type": "string"},
},
"required": []string{"steps", "final_answer"},
"additionalProperties": false,
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.chat.completions.ChatCompletionCreateParams;
import java.util.List;
import java.util.Map;
Map<String, Object> stepSchema =
Map.of(
"type",
"object",
"properties",
Map.of(
"explanation", Map.of("type", "string"),
"output", Map.of("type", "string")),
"required",
List.of("explanation", "output"),
"additionalProperties",
false);
Map<String, Object> mathSchema =
Map.of(
"type",
"object",
"properties",
Map.of(
"steps", Map.of("type", "array", "items", stepSchema),
"final_answer", Map.of("type", "string")),
"required",
List.of("steps", "final_answer"),
"additionalProperties",
false);
ChatCompletionCreateParams params =
ChatCompletionCreateParams.builder()
.model("gpt-6-astra")
.addSystemMessage(
"You are a helpful math tutor. Guide the user through the solution step by step.")
.addUserMessage("How can I solve 8x + 7 = -23?")
.store(true)
.putAdditionalBodyProperty(
"response_format",
JsonValue.from(
Map.of(
"type",
"json_schema",
"json_schema",
Map.of("name", "math_response", "strict", true, "schema", mathSchema))))
.build();
client.chat().completions().create(params).choices().stream()
.flatMap(choice -> choice.message().content().stream())
.forEach(System.out::println);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46using System.Text.Json;
using OpenAI.Chat;
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
string model = "gpt-6-astra";
ChatClient client = new(model, key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": { "type": "string" },
"output": { "type": "string" }
},
"required": ["explanation", "output"],
"additionalProperties": false
}
},
"final_answer": { "type": "string" }
},
"required": ["steps", "final_answer"],
"additionalProperties": false
}
"""
);
ChatCompletionOptions options = new()
{
ResponseFormat = ChatResponseFormat.CreateJsonSchemaFormat(
"math_response",
schema,
jsonSchemaIsStrict: true
),
};
ChatCompletion completion = await client.CompleteChatAsync(
[new SystemChatMessage("You are a helpful math tutor. Guide the user through the solution step by step."), new UserChatMessage("How can I solve 8x + 7 = -23?")],
options
);
using JsonDocument parsed = JsonDocument.Parse(completion.Content[0].Text);
Console.WriteLine(parsed.RootElement);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48require "openai"
client = OpenAI::Client.new
math_schema = {
type: :object,
properties: {
steps: {
type: :array,
items: {
type: :object,
properties: {
explanation: { type: :string },
output: { type: :string }
},
required: %w[explanation output],
additionalProperties: false
}
},
final_answer: { type: :string }
},
required: %w[steps final_answer],
additionalProperties: false
}
completion = client.chat.completions.create(
model: "gpt-6-astra",
messages: [
{
role: :system,
content: "You are a helpful math tutor. Guide the user through the solution step by step."
},
{
role: :user,
content: "How can I solve 8x + 7 = -23?"
}
],
store: true,
response_format: {
type: :json_schema,
json_schema: {
name: "math_response",
strict: true,
schema: math_schema
}
}
)
puts(completion.choices.fetch(0).message.content)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43curl https://api.openai.com/v1/chat/completions \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6-astra",
"messages": [
{
"role": "system",
"content": "You are a helpful math tutor. Guide the user through the solution step by step."
},
{
"role": "user",
"content": "how can I solve 8x + 7 = -23"
}
],
"response_format": {
"type": "json_schema",
"json_schema": {
"name": "math_response",
"schema": {
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": { "type": "string" },
"output": { "type": "string" }
},
"required": ["explanation", "output"],
"additionalProperties": false
}
},
"final_answer": { "type": "string" }
},
"required": ["steps", "final_answer"],
"additionalProperties": false
},
"strict": true
}
}
}'1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40const response = await openai.responses.create({
model: "gpt-6-astra",
input: [
{
role: "system",
content:
"You are a helpful math tutor. Guide the user through the solution step by step.",
},
{ role: "user", content: "how can I solve 8x + 7 = -23" },
],
text: {
format: {
type: "json_schema",
name: "math_response",
schema: {
type: "object",
properties: {
steps: {
type: "array",
items: {
type: "object",
properties: {
explanation: { type: "string" },
output: { type: "string" },
},
required: ["explanation", "output"],
additionalProperties: false,
},
},
final_answer: { type: "string" },
},
required: ["steps", "final_answer"],
additionalProperties: false,
},
strict: true,
},
},
});
console.log(response.output_text);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39response = client.responses.create(
model="gpt-6-astra",
input=[
{
"role": "system",
"content": "You are a helpful math tutor. Guide the user through the solution step by step.",
},
{"role": "user", "content": "how can I solve 8x + 7 = -23"},
],
text={
"format": {
"type": "json_schema",
"name": "math_response",
"schema": {
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": {"type": "string"},
"output": {"type": "string"},
},
"required": ["explanation", "output"],
"additionalProperties": False,
},
},
"final_answer": {"type": "string"},
},
"required": ["steps", "final_answer"],
"additionalProperties": False,
},
"strict": True,
},
},
)
print(response.output_text)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/responses"
)
func main() {
client := openai.NewClient()
response, err := client.Responses.New(context.Background(), responses.ResponseNewParams{
Model: "gpt-6-astra",
Input: responses.ResponseNewParamsInputUnion{OfInputItemList: responses.ResponseInputParam{
responses.ResponseInputItemParamOfMessage(
responses.ResponseInputMessageContentListParam{responses.ResponseInputContentParamOfInputText("You are a helpful math tutor. Guide the user through the solution step by step.")},
responses.EasyInputMessageRoleSystem,
),
responses.ResponseInputItemParamOfMessage(
responses.ResponseInputMessageContentListParam{responses.ResponseInputContentParamOfInputText("how can I solve 8x + 7 = -23")},
responses.EasyInputMessageRoleUser,
),
}},
Text: responses.ResponseTextConfigParam{Format: responses.ResponseFormatTextConfigUnionParam{
OfJSONSchema: &responses.ResponseFormatTextJSONSchemaConfigParam{Name: "math_response", Schema: mathSchema(), Strict: openai.Bool(true)},
}},
})
if err != nil {
panic(err)
}
fmt.Println(response.OutputText())
}
func mathSchema() map[string]any {
return map[string]any{
"type": "object",
"properties": map[string]any{
"steps": map[string]any{"type": "array", "items": map[string]any{"type": "object", "properties": map[string]any{"explanation": map[string]any{"type": "string"}, "output": map[string]any{"type": "string"}}, "required": []string{"explanation", "output"}, "additionalProperties": false}},
"final_answer": map[string]any{"type": "string"},
},
"required": []string{"steps", "final_answer"},
"additionalProperties": false,
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.responses.EasyInputMessage;
import com.openai.models.responses.ResponseCreateParams;
import com.openai.models.responses.ResponseFormatTextJsonSchemaConfig;
import com.openai.models.responses.ResponseInputItem;
import com.openai.models.responses.ResponseTextConfig;
import java.util.List;
import java.util.Map;
Map<String, Object> schema =
Map.of(
"type",
"object",
"properties",
Map.of(
"steps",
Map.of(
"type",
"array",
"items",
Map.of(
"type",
"object",
"properties",
Map.of(
"explanation", Map.of("type", "string"),
"output", Map.of("type", "string")),
"required",
List.of("explanation", "output"),
"additionalProperties",
false)),
"final_answer", Map.of("type", "string")),
"required",
List.of("steps", "final_answer"),
"additionalProperties",
false);
ResponseCreateParams params =
ResponseCreateParams.builder()
.model("gpt-6-astra")
.inputOfResponse(
List.of(
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.SYSTEM)
.content(
"You are a helpful math tutor. Guide the user through the solution step by step.")
.build()),
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.USER)
.content("How can I solve 8x + 7 = -23?")
.build())))
.text(
ResponseTextConfig.builder()
.format(
ResponseFormatTextJsonSchemaConfig.builder()
.name("math_response")
.strict(true)
.schema(
JsonValue.from(schema)
.convert(ResponseFormatTextJsonSchemaConfig.Schema.class))
.build())
.build())
.build();
client.responses().create(params).output().stream()
.flatMap(item -> item.message().stream())
.flatMap(message -> message.content().stream())
.flatMap(content -> content.outputText().stream())
.forEach(text -> System.out.println(text.text()));1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49using System.Text.Json;
using OpenAI.Responses;
#pragma warning disable OPENAI001
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ResponsesClient client = new(key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": { "type": "string" },
"output": { "type": "string" }
},
"required": ["explanation", "output"],
"additionalProperties": false
}
},
"final_answer": { "type": "string" }
},
"required": ["steps", "final_answer"],
"additionalProperties": false
}
"""
);
CreateResponseOptions options = new()
{
Model = "gpt-6-astra",
TextOptions = new ResponseTextOptions
{
TextFormat = ResponseTextFormat.CreateJsonSchemaFormat(
"math_response",
schema,
jsonSchemaIsStrict: true
),
},
};
options.InputItems.Add(ResponseItem.CreateSystemMessageItem("You are a helpful math tutor. Guide the user through the solution step by step."));
options.InputItems.Add(ResponseItem.CreateUserMessageItem("How can I solve 8x + 7 = -23?"));
ResponseResult response = await client.CreateResponseAsync(options);
using JsonDocument parsed = JsonDocument.Parse(response.GetOutputText());
Console.WriteLine(parsed.RootElement);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47require "openai"
client = OpenAI::Client.new
math_schema = {
type: :object,
properties: {
steps: {
type: :array,
items: {
type: :object,
properties: {
explanation: { type: :string },
output: { type: :string }
},
required: %w[explanation output],
additionalProperties: false
}
},
final_answer: { type: :string }
},
required: %w[steps final_answer],
additionalProperties: false
}
response = client.responses.create(
model: "gpt-6-astra",
input: [
{
role: :system,
content: "You are a helpful math tutor. Guide the user through the solution step by step."
},
{
role: :user,
content: "How can I solve 8x + 7 = -23?"
}
],
text: {
format: {
type: :json_schema,
name: "math_response",
strict: true,
schema: math_schema
}
}
)
puts(response.output_text)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43curl https://api.openai.com/v1/responses \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6-astra",
"input": [
{
"role": "system",
"content": "You are a helpful math tutor. Guide the user through the solution step by step."
},
{
"role": "user",
"content": "how can I solve 8x + 7 = -23"
}
],
"text": {
"format": {
"type": "json_schema",
"name": "math_response",
"schema": {
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": { "type": "string" },
"output": { "type": "string" }
},
"required": ["explanation", "output"],
"additionalProperties": false
}
},
"final_answer": { "type": "string" }
},
"required": ["steps", "final_answer"],
"additionalProperties": false
},
"strict": true
}
}
}'참고: 각 스키마를 처음 사용하는 요청에서는 API가 해당 스키마를 처리하는 데 시간이 더 걸립니다. 이후 같은 스키마를 사용하는 요청에서는 이로 인한 추가 지연이 없습니다.
모델이 지정된 JSON 스키마에 맞는 유효한 응답을 생성하지 못할 때가 있습니다.
예를 들어 모델이 안전상의 이유로 답변을 거부하거나 토큰 수가 한도에 도달해 응답이 중간에 끊기는 경우입니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70try {
const completion = await openai.chat.completions.create({
model: "gpt-6-astra",
messages: [
{
role: "system",
content:
"You are a helpful math tutor. Guide the user through the solution step by step.",
},
{
role: "user",
content: "how can I solve 8x + 7 = -23",
},
],
store: true,
response_format: {
type: "json_schema",
json_schema: {
name: "math_response",
schema: {
type: "object",
properties: {
steps: {
type: "array",
items: {
type: "object",
properties: {
explanation: {
type: "string",
},
output: {
type: "string",
},
},
required: ["explanation", "output"],
additionalProperties: false,
},
},
final_answer: {
type: "string",
},
},
required: ["steps", "final_answer"],
additionalProperties: false,
},
strict: true,
},
},
max_completion_tokens: 1024,
});
if (completion.choices[0].finish_reason === "length") {
// Handle the case where the model did not return a complete response
throw new Error("Incomplete response");
}
const math_response = completion.choices[0].message;
if (math_response.refusal) {
// handle refusal
console.log(math_response.refusal);
} else if (math_response.content) {
console.log(math_response.content);
} else {
throw new Error("No response content");
}
} catch (e) {
// Handle edge cases
console.error(e);
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54try:
response = client.chat.completions.create(
model="gpt-6-astra",
messages=[
{
"role": "system",
"content": "You are a helpful math tutor. Guide the user through the solution step by step.",
},
{"role": "user", "content": "how can I solve 8x + 7 = -23"},
],
response_format={
"type": "json_schema",
"json_schema": {
"name": "math_response",
"strict": True,
"schema": {
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": {"type": "string"},
"output": {"type": "string"},
},
"required": ["explanation", "output"],
"additionalProperties": False,
},
},
"final_answer": {"type": "string"},
},
"required": ["steps", "final_answer"],
"additionalProperties": False,
},
},
},
max_completion_tokens=1024,
)
if response.choices[0].finish_reason == "length":
raise Exception("Incomplete response")
math_response = response.choices[0].message
if math_response.refusal:
print(math_response.refusal)
elif math_response.content:
print(math_response.content)
else:
raise Exception("No response content")
except Exception as e:
# handle errors like finish_reason, refusal, content_filter, etc.
print(e)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56package main
import (
"context"
"errors"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/shared"
)
func main() {
client := openai.NewClient()
completion, err := client.Chat.Completions.New(context.Background(), openai.ChatCompletionNewParams{
Model: "gpt-6-astra",
Messages: []openai.ChatCompletionMessageParamUnion{
openai.SystemMessage("You are a helpful math tutor. Guide the user through the solution step by step."),
openai.UserMessage("how can I solve 8x + 7 = -23"),
},
Store: openai.Bool(true),
MaxCompletionTokens: openai.Int(1024),
ResponseFormat: openai.ChatCompletionNewParamsResponseFormatUnion{
OfJSONSchema: &shared.ResponseFormatJSONSchemaParam{JSONSchema: shared.ResponseFormatJSONSchemaJSONSchemaParam{
Name: "math_response", Schema: mathSchema(), Strict: openai.Bool(true),
}},
},
})
if err != nil {
panic(err)
}
choice := completion.Choices[0]
if choice.FinishReason == "length" {
panic(errors.New("incomplete response"))
}
if choice.Message.Refusal != "" {
fmt.Println(choice.Message.Refusal)
return
}
if choice.Message.Content == "" {
panic(errors.New("no response content"))
}
fmt.Println(choice.Message.Content)
}
func mathSchema() map[string]any {
return map[string]any{
"type": "object",
"properties": map[string]any{
"steps": map[string]any{"type": "array", "items": map[string]any{"type": "object", "properties": map[string]any{"explanation": map[string]any{"type": "string"}, "output": map[string]any{"type": "string"}}, "required": []string{"explanation", "output"}, "additionalProperties": false}},
"final_answer": map[string]any{"type": "string"},
},
"required": []string{"steps", "final_answer"},
"additionalProperties": false,
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.chat.completions.ChatCompletion;
import com.openai.models.chat.completions.ChatCompletionCreateParams;
import java.util.List;
import java.util.Map;
Map<String, Object> stepSchema =
Map.of(
"type",
"object",
"properties",
Map.of(
"explanation", Map.of("type", "string"),
"output", Map.of("type", "string")),
"required",
List.of("explanation", "output"),
"additionalProperties",
false);
Map<String, Object> mathSchema =
Map.of(
"type",
"object",
"properties",
Map.of(
"steps", Map.of("type", "array", "items", stepSchema),
"final_answer", Map.of("type", "string")),
"required",
List.of("steps", "final_answer"),
"additionalProperties",
false);
ChatCompletionCreateParams params =
ChatCompletionCreateParams.builder()
.model("gpt-6-astra")
.addSystemMessage(
"You are a helpful math tutor. Guide the user through the solution step by step.")
.addUserMessage("How can I solve 8x + 7 = -23?")
.maxCompletionTokens(1024)
.store(true)
.putAdditionalBodyProperty(
"response_format",
JsonValue.from(
Map.of(
"type",
"json_schema",
"json_schema",
Map.of("name", "math_response", "strict", true, "schema", mathSchema))))
.build();
var choice = client.chat().completions().create(params).choices().get(0);
if (choice.finishReason().equals(ChatCompletion.Choice.FinishReason.LENGTH)) {
System.out.println("Incomplete response");
} else if (choice.message().refusal().isPresent()) {
System.out.println(choice.message().refusal().orElseThrow());
} else {
System.out.println(
choice
.message()
.content()
.orElseThrow(() -> new IllegalStateException("No response content")));
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64using OpenAI.Chat;
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
string model = "gpt-6-astra";
ChatClient client = new(model, key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": { "type": "string" },
"output": { "type": "string" }
},
"required": ["explanation", "output"],
"additionalProperties": false
}
},
"final_answer": { "type": "string" }
},
"required": ["steps", "final_answer"],
"additionalProperties": false
}
"""
);
ChatCompletionOptions options = new()
{
MaxOutputTokenCount = 300,
ResponseFormat = ChatResponseFormat.CreateJsonSchemaFormat(
"math_response",
schema,
jsonSchemaIsStrict: true
),
};
ChatCompletion completion = await client.CompleteChatAsync(
[new SystemChatMessage("You are a helpful math tutor. Guide the user through the solution step by step."), new UserChatMessage("How can I solve 8x + 7 = -23?")],
options
);
if (completion.FinishReason == ChatFinishReason.Length)
{
throw new InvalidOperationException("The structured response was incomplete.");
}
if (completion.FinishReason == ChatFinishReason.ContentFilter)
{
throw new InvalidOperationException("The structured response was interrupted by the content filter.");
}
if (!string.IsNullOrEmpty(completion.Refusal))
{
Console.WriteLine(completion.Refusal);
}
else if (completion.Content.Count > 0)
{
Console.WriteLine(completion.Content[0].Text);
}
else
{
throw new InvalidOperationException("The completion did not contain a response.");
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58require "openai"
client = OpenAI::Client.new
step_schema = {
type: :object,
properties: {
explanation: { type: :string },
output: { type: :string }
},
required: %w[explanation output],
additionalProperties: false
}
math_schema = {
type: :object,
properties: {
steps: {
type: :array,
items: step_schema
},
final_answer: { type: :string }
},
required: %w[steps final_answer],
additionalProperties: false
}
completion = client.chat.completions.create(
model: "gpt-6-astra",
messages: [
{
role: :system,
content: "You are a helpful math tutor. Guide the user through the solution step by step."
},
{
role: :user,
content: "How can I solve 8x + 7 = -23?"
}
],
max_completion_tokens: 1_024,
store: true,
response_format: {
type: :json_schema,
json_schema: {
name: "math_response",
strict: true,
schema: math_schema
}
}
)
choice = completion.choices.fetch(0)
if choice.finish_reason == OpenAI::Chat::ChatCompletion::Choice::FinishReason::LENGTH
raise "Incomplete response"
elsif choice.message.refusal
puts(choice.message.refusal)
else
content = choice.message.content or raise "No response content"
puts(content)
end1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77try {
const response = await openai.responses.create({
model: "gpt-6-astra",
input: [
{
role: "system",
content:
"You are a helpful math tutor. Guide the user through the solution step by step.",
},
{
role: "user",
content: "how can I solve 8x + 7 = -23",
},
],
max_output_tokens: 1024,
text: {
format: {
type: "json_schema",
name: "math_response",
schema: {
type: "object",
properties: {
steps: {
type: "array",
items: {
type: "object",
properties: {
explanation: {
type: "string",
},
output: {
type: "string",
},
},
required: ["explanation", "output"],
additionalProperties: false,
},
},
final_answer: {
type: "string",
},
},
required: ["steps", "final_answer"],
additionalProperties: false,
},
strict: true,
},
},
});
if (
response.status === "incomplete" &&
response.incomplete_details.reason === "max_output_tokens"
) {
// Handle the case where the model did not return a complete response
throw new Error("Incomplete response");
}
const message = response.output.find((item) => item.type === "message");
const math_response = message?.content[0];
if (!math_response) {
throw new Error("No response content");
}
if (math_response.type === "refusal") {
// handle refusal
console.log(math_response.refusal);
} else if (math_response.type === "output_text") {
console.log(math_response.text);
} else {
throw new Error("No response content");
}
} catch (e) {
// Handle edge cases
console.error(e);
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61try:
response = client.responses.create(
model="gpt-6-astra",
input=[
{
"role": "system",
"content": "You are a helpful math tutor. Guide the user through the solution step by step.",
},
{"role": "user", "content": "how can I solve 8x + 7 = -23"},
],
text={
"format": {
"type": "json_schema",
"name": "math_response",
"strict": True,
"schema": {
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": {"type": "string"},
"output": {"type": "string"},
},
"required": ["explanation", "output"],
"additionalProperties": False,
},
},
"final_answer": {"type": "string"},
},
"required": ["steps", "final_answer"],
"additionalProperties": False,
},
},
},
max_output_tokens=1024,
)
if (
response.status == "incomplete"
and response.incomplete_details.reason == "max_output_tokens"
):
raise Exception("Incomplete response")
message = next((item for item in response.output if item.type == "message"), None)
math_response = message.content[0] if message and message.content else None
if not math_response:
raise Exception("No response content")
if math_response.type == "refusal":
print(math_response.refusal)
elif math_response.type == "output_text":
print(math_response.text)
else:
raise Exception("No response content")
except Exception as e:
# handle errors like finish_reason, refusal, content_filter, etc.
print(e)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66package main
import (
"context"
"errors"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/responses"
)
func main() {
client := openai.NewClient()
response, err := client.Responses.New(context.Background(), responses.ResponseNewParams{
Model: "gpt-6-astra",
Input: responses.ResponseNewParamsInputUnion{OfInputItemList: responses.ResponseInputParam{
responses.ResponseInputItemParamOfMessage(
responses.ResponseInputMessageContentListParam{responses.ResponseInputContentParamOfInputText("You are a helpful math tutor. Guide the user through the solution step by step.")},
responses.EasyInputMessageRoleSystem,
),
responses.ResponseInputItemParamOfMessage(
responses.ResponseInputMessageContentListParam{responses.ResponseInputContentParamOfInputText("how can I solve 8x + 7 = -23")},
responses.EasyInputMessageRoleUser,
),
}},
MaxOutputTokens: openai.Int(1024),
Text: responses.ResponseTextConfigParam{Format: responses.ResponseFormatTextConfigUnionParam{
OfJSONSchema: &responses.ResponseFormatTextJSONSchemaConfigParam{Name: "math_response", Schema: mathSchema(), Strict: openai.Bool(true)},
}},
})
if err != nil {
panic(err)
}
if response.Status == "incomplete" {
panic(errors.New("incomplete response"))
}
for _, output := range response.Output {
if output.Type != "message" {
continue
}
for _, content := range output.AsMessage().Content {
if content.Type == "refusal" {
fmt.Println(content.AsRefusal().Refusal)
return
}
if content.Type == "output_text" {
fmt.Println(content.AsOutputText().Text)
return
}
}
}
panic(errors.New("no response content"))
}
func mathSchema() map[string]any {
return map[string]any{
"type": "object",
"properties": map[string]any{
"steps": map[string]any{"type": "array", "items": map[string]any{"type": "object", "properties": map[string]any{"explanation": map[string]any{"type": "string"}, "output": map[string]any{"type": "string"}}, "required": []string{"explanation", "output"}, "additionalProperties": false}},
"final_answer": map[string]any{"type": "string"},
},
"required": []string{"steps", "final_answer"},
"additionalProperties": false,
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.responses.EasyInputMessage;
import com.openai.models.responses.ResponseCreateParams;
import com.openai.models.responses.ResponseFormatTextJsonSchemaConfig;
import com.openai.models.responses.ResponseInputItem;
import com.openai.models.responses.ResponseStatus;
import com.openai.models.responses.ResponseTextConfig;
import java.util.List;
import java.util.Map;
ResponseCreateParams params =
ResponseCreateParams.builder()
.model("gpt-6-astra")
.inputOfResponse(
List.of(
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.SYSTEM)
.content(
"You are a helpful math tutor. Guide the user through the solution step by step.")
.build()),
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.USER)
.content("How can I solve 8x + 7 = -23?")
.build())))
.text(
ResponseTextConfig.builder()
.format(
ResponseFormatTextJsonSchemaConfig.builder()
.name("math_response")
.strict(true)
.schema(
ResponseFormatTextJsonSchemaConfig.Schema.builder()
.putAdditionalProperty("type", JsonValue.from("object"))
.putAdditionalProperty(
"properties",
JsonValue.from(
Map.of(
"steps",
Map.of(
"type",
"array",
"items",
Map.of(
"type",
"object",
"properties",
Map.of(
"explanation",
Map.of("type", "string"),
"output",
Map.of("type", "string")),
"required",
List.of("explanation", "output"),
"additionalProperties",
false)),
"final_answer",
Map.of("type", "string"))))
.putAdditionalProperty(
"required",
JsonValue.from(List.of("steps", "final_answer")))
.putAdditionalProperty(
"additionalProperties", JsonValue.from(false))
.build())
.build())
.build())
.maxOutputTokens(1_024L)
.build();
var response = client.responses().create(params);
if (response.status().filter(ResponseStatus.INCOMPLETE::equals).isPresent()) {
throw new IllegalStateException("Incomplete response");
}
var content =
response.output().stream()
.flatMap(item -> item.message().stream())
.flatMap(message -> message.content().stream())
.findFirst()
.orElseThrow(() -> new IllegalStateException("No response content"));
if (content.refusal().isPresent()) {
System.out.println(content.refusal().orElseThrow().refusal());
} else {
System.out.println(
content
.outputText()
.orElseThrow(() -> new IllegalStateException("No response content"))
.text());
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68using OpenAI.Responses;
#pragma warning disable OPENAI001
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ResponsesClient client = new(key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": { "type": "string" },
"output": { "type": "string" }
},
"required": ["explanation", "output"],
"additionalProperties": false
}
},
"final_answer": { "type": "string" }
},
"required": ["steps", "final_answer"],
"additionalProperties": false
}
"""
);
CreateResponseOptions options = new()
{
Model = "gpt-6-astra",
MaxOutputTokenCount = 300,
TextOptions = new ResponseTextOptions
{
TextFormat = ResponseTextFormat.CreateJsonSchemaFormat(
"math_response",
schema,
jsonSchemaIsStrict: true
),
},
};
options.InputItems.Add(ResponseItem.CreateSystemMessageItem("You are a helpful math tutor. Guide the user through the solution step by step."));
options.InputItems.Add(ResponseItem.CreateUserMessageItem("How can I solve 8x + 7 = -23?"));
ResponseResult response = await client.CreateResponseAsync(options);
if (
response.Status == ResponseStatus.Incomplete
&& response.IncompleteStatusDetails?.Reason == ResponseIncompleteStatusReason.MaxOutputTokens
)
{
throw new InvalidOperationException("The structured response was incomplete.");
}
if (
response.Status == ResponseStatus.Incomplete
&& response.IncompleteStatusDetails?.Reason == ResponseIncompleteStatusReason.ContentFilter
)
{
throw new InvalidOperationException("The structured response was interrupted by the content filter.");
}
MessageResponseItem message = response.OutputItems.OfType<MessageResponseItem>().FirstOrDefault()
?? throw new InvalidOperationException("The response did not include an output message.");
ResponseContentPart content = message.Content.FirstOrDefault()
?? throw new InvalidOperationException("The response did not include output content.");
Console.WriteLine(
content.Kind == ResponseContentPartKind.Refusal ? content.Refusal : content.Text
);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65require "openai"
client = OpenAI::Client.new
step_schema = {
type: :object,
properties: {
explanation: { type: :string },
output: { type: :string }
},
required: %w[explanation output],
additionalProperties: false
}
math_schema = {
type: :object,
properties: {
steps: {
type: :array,
items: step_schema
},
final_answer: { type: :string }
},
required: %w[steps final_answer],
additionalProperties: false
}
response = client.responses.create(
model: "gpt-6-astra",
input: [
{
role: :system,
content: "You are a helpful math tutor. Guide the user through the solution step by step."
},
{
role: :user,
content: "How can I solve 8x + 7 = -23?"
}
],
max_output_tokens: 1_024,
text: {
format: {
type: :json_schema,
name: "math_response",
strict: true,
schema: math_schema
}
}
)
if response.status == OpenAI::Responses::ResponseStatus::INCOMPLETE
raise "Incomplete response"
end
message = response.output.find do |item|
item.is_a?(OpenAI::Models::Responses::ResponseOutputMessage)
end
unless message.is_a?(OpenAI::Models::Responses::ResponseOutputMessage)
raise "No response message"
end
content = message.content.fetch(0)
if content.is_a?(OpenAI::Models::Responses::ResponseOutputRefusal)
puts(content.refusal)
else
puts(content.text)
end응답에 스키마와 일치하는 JSON이 포함되어 있는지 확인한 뒤, 해당 JSON을 파싱해 사용 중인 언어의 기본 데이터 구조로 변환하세요. 타입을 지원하는 언어에서는 데이터 구조에 맞는 타입이나 클래스로 데이터를 모델링할 수도 있습니다.
다음은 예시입니다.
1
2
3
4
5// The request that produces `response` appears earlier in this guide.
const content = response.choices[0].message.content;
if (!content) throw new Error("The response did not contain JSON output.");
const solution = JSON.parse(content);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18from pydantic import BaseModel, ValidationError
class Step(BaseModel):
explanation: str
output: str
class Solution(BaseModel):
steps: list[Step]
final_answer: str
try:
solution = Solution.model_validate_json(response.choices[0].message.content)
print(solution)
except ValidationError as error:
print(error.json())1
2
3
4
5
6
7
8
9
10
11var solution struct {
Steps []struct {
Explanation string `json:"explanation"`
Output string `json:"output"`
} `json:"steps"`
FinalAnswer string `json:"final_answer"`
}
if err := json.Unmarshal([]byte(completion.Choices[0].Message.Content), &solution); err != nil {
return err
}
fmt.Println(solution)1System.out.println(new ObjectMapper().readTree(content));1
2
3
4
5using System.Text.Json;
using OpenAI.Chat;
using JsonDocument solution = JsonDocument.Parse(completion.Content[0].Text);
Console.WriteLine(solution.RootElement);1
2
3
4
5content = completion.choices.fetch(0).message.content
raise "Response did not contain JSON output." if content.nil?
solution = JSON.parse(content)
puts(solution)대화 도중 응답 형식 변경하기
다음 응답부터 새 스키마를 적용하려면 configuration_update 항목을 input에 추가한 뒤 다음 사용자 메시지를 추가하세요. 요청마다 최상위 text.format에 원래 값을 유지하면 이전 프롬프트의 접두부가 계속 프롬프트 캐싱 대상이 됩니다.
GPT-6 및 이후 모델 중 이 기능을 지원하는 모델에서는 표준 단일 에이전트 모드로 응답 형식을 변경할 수 있습니다.
이전 턴에서 빨간색 상자 17개와 파란색 상자 26개로 구성된 화물을 일반 텍스트로 설명했다고 가정해 보겠습니다. previous_response_id로 대화를 이어가고 다음 배열을 input으로 전달하세요.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28import OpenAI from "openai";
import { zodTextFormat } from "openai/helpers/zod";
import { z } from "zod";
const client = new OpenAI();
const model = "gpt-6-astra";
const CrateCount = z.object({ total_crates: z.number().int() });
const first = await client.responses.create({
model,
input:
"Summarize this shipment in one sentence: 17 red crates and 26 blue crates.",
store: true,
});
const format = zodTextFormat(CrateCount, "crate_count");
const second = await client.responses.parse({
model,
previous_response_id: first.id,
input: [
{ type: "configuration_update", text: { format } },
{ role: "user", content: "How many crates are in the shipment?" },
],
});
if (second.status === "completed" && second.output_parsed) {
console.log(second.output_parsed.total_crates);
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40from openai import OpenAI
from pydantic import BaseModel, ConfigDict
class CrateCount(BaseModel):
model_config = ConfigDict(extra="forbid")
total_crates: int
client = OpenAI()
model = "gpt-6-astra"
first = client.responses.create(
model=model,
input="Summarize this shipment in one sentence: 17 red crates and 26 blue crates.",
store=True,
)
second = client.responses.parse(
model=model,
previous_response_id=first.id,
input=[
{
"type": "configuration_update",
"text": {
"format": {
"type": "json_schema",
"name": "crate_count",
"schema": CrateCount.model_json_schema(),
"strict": True,
}
},
},
{"role": "user", "content": "How many crates are in the shipment?"},
],
parse_as=CrateCount,
)
if second.status == "completed" and second.output_parsed is not None:
print(second.output_parsed.total_crates)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/responses"
)
func main() {
client := openai.NewClient()
ctx := context.Background()
const model = "gpt-6-astra"
first, err := client.Responses.New(ctx, responses.ResponseNewParams{
Model: model,
Input: responses.ResponseNewParamsInputUnion{
OfString: openai.String("Summarize this shipment in one sentence: 17 red crates and 26 blue crates."),
},
Store: openai.Bool(true),
})
if err != nil {
panic(err)
}
format := responses.ResponseFormatTextJSONSchemaConfigParam{
Name: "crate_count",
Schema: map[string]any{
"type": "object",
"properties": map[string]any{"total_crates": map[string]any{"type": "integer"}},
"required": []string{"total_crates"},
"additionalProperties": false,
},
Strict: openai.Bool(true),
}
second, err := client.Responses.New(ctx, responses.ResponseNewParams{
Model: model,
PreviousResponseID: openai.String(first.ID),
Input: responses.ResponseNewParamsInputUnion{OfInputItemList: responses.ResponseInputParam{
{OfConfigurationUpdate: &responses.ResponseConfigurationUpdateItemParam{
Text: responses.ResponseConfigurationUpdateItemParamText{
Format: responses.ResponseConfigurationUpdateItemParamTextFormatUnion{OfJSONSchema: &format},
},
}},
responses.ResponseInputItemParamOfMessage("How many crates are in the shipment?", responses.EasyInputMessageRoleUser),
}},
})
if err != nil {
panic(err)
}
if second.Status != responses.ResponseStatusCompleted {
return
}
for _, item := range second.Output {
if item.Type != "message" || (item.Phase != "" && item.Phase != responses.ResponseOutputMessagePhaseFinalAnswer) {
continue
}
for _, part := range item.Content {
if part.Type == "output_text" {
fmt.Println(part.Text)
}
}
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73import com.openai.core.JsonValue;
import com.openai.models.responses.EasyInputMessage;
import com.openai.models.responses.ResponseConfigurationUpdateItemParam;
import com.openai.models.responses.ResponseCreateParams;
import com.openai.models.responses.ResponseFormatTextJsonSchemaConfig;
import com.openai.models.responses.ResponseInputItem;
import com.openai.models.responses.ResponseOutputMessage;
import com.openai.models.responses.ResponseStatus;
import java.util.List;
import java.util.Map;
String model = "gpt-6-astra";
var first =
client
.responses()
.create(
ResponseCreateParams.builder()
.model(model)
.input(
"Summarize this shipment in one sentence: 17 red crates and 26 blue crates.")
.store(true)
.build());
var schema =
Map.of(
"type",
"object",
"properties",
Map.of("total_crates", Map.of("type", "integer")),
"required",
List.of("total_crates"),
"additionalProperties",
false);
var format =
ResponseFormatTextJsonSchemaConfig.builder()
.name("crate_count")
.schema(JsonValue.from(schema).convert(ResponseFormatTextJsonSchemaConfig.Schema.class))
.strict(true)
.build();
var update =
ResponseConfigurationUpdateItemParam.builder()
.text(ResponseConfigurationUpdateItemParam.Text.builder().format(format).build())
.build();
var question =
EasyInputMessage.builder()
.role(EasyInputMessage.Role.USER)
.content("How many crates are in the shipment?")
.build();
var second =
client
.responses()
.create(
ResponseCreateParams.builder()
.model(model)
.previousResponseId(first.id())
.inputOfResponse(
List.of(
ResponseInputItem.ofConfigurationUpdate(update),
ResponseInputItem.ofEasyInputMessage(question)))
.build());
if (!ResponseStatus.COMPLETED.equals(second.status().orElse(null))) {
return;
}
second.output().stream()
.flatMap(item -> item.message().stream())
.filter(
message ->
message.phase().map(ResponseOutputMessage.Phase.FINAL_ANSWER::equals).orElse(true))
.flatMap(message -> message.content().stream())
.flatMap(content -> content.outputText().stream())
.forEach(text -> System.out.println(text.text()));1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52# gem install openai sorbet-runtime
require "openai"
require "openai/helpers/sorbet"
class CrateCount < T::Struct
const :total_crates, Integer
end
client = OpenAI::Client.new
model = "gpt-6-astra"
first = client.responses.create(
model: model,
input: "Summarize this shipment in one sentence: 17 red crates and 26 blue crates.",
store: true
)
second = client.responses.create(
model: model,
previous_response_id: first.id,
input: [
{
type: :configuration_update,
text: {
format: {
type: :json_schema,
name: "crate_count",
schema: OpenAI::StructuredOutput.from_sorbet(CrateCount),
strict: true
}
}
},
{
role: :user,
content: "How many crates are in the shipment?"
}
]
)
if second.status == OpenAI::Responses::ResponseStatus::COMPLETED
second.output.grep(OpenAI::Responses::ResponseOutputMessage).each do |message|
next unless message.phase.nil? || message.phase == :final_answer
message.content.grep(OpenAI::Responses::ResponseOutputText).each do |part|
count = part.parsed
case count
when CrateCount
puts count.total_crates
end
end
end
end예상 답변은 {"total_crates":43}입니다. 응답 객체의 text.format에는 여전히 요청의 최상위 수준에서 지정한 원래 형식이 표시됩니다. 모델이 요청을 거부했는지, 응답이 불완전한지 확인한 뒤 가장 최근의 형식 업데이트에 지정한 스키마로 답변을 파싱하세요.
형식을 업데이트할 때마다 기존 형식 전체가 교체됩니다. 새 형식은 이후 턴에도 계속 적용되며, reasoning.effort만 변경해도 그대로 유지됩니다. 일반 텍스트로 돌아가려면 {"type":"configuration_update","text":{"format":{"type":"text"}}}를 추가하세요. 업데이트에서 지정할 수 있는 형식은 json_schema와 text뿐입니다.
대화 기록을 직접 다시 전송할 때는 각 업데이트를 원래 위치에 두세요. 컨텍스트 압축을 명시적으로 실행한 후에는 변경했던 설정을 각각 다시 지정하세요. 자세한 내용은 설정 업데이트 호환성을 참고하세요.
구조화된 출력값의 거부 응답
사용자가 입력한 내용을 구조화된 출력값으로 처리할 때 OpenAI 모델은 간혹 안전상의 이유로 요청을 거부할 수 있습니다. 거부 응답은 response_format에 지정한 스키마를 따르지 않을 수 있으므로, API 응답에는 모델이 요청을 거부했음을 나타내는 refusal이라는 새 필드가 포함됩니다.
출력 객체에 refusal 속성이 있으면 UI에 거부 응답을 표시하거나, 응답을 처리하는 코드에 조건문을 추가해 거부된 요청을 별도로 처리할 수 있습니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31const Step = z.object({
explanation: z.string(),
output: z.string(),
});
const MathReasoning = z.object({
steps: z.array(Step),
final_answer: z.string(),
});
const completion = await openai.chat.completions.parse({
model: "gpt-6-astra",
messages: [
{
role: "system",
content:
"You are a helpful math tutor. Guide the user through the solution step by step.",
},
{ role: "user", content: "how can I solve 8x + 7 = -23" },
],
response_format: zodResponseFormat(MathReasoning, "math_reasoning"),
});
const math_reasoning = completion.choices[0].message;
// If the model refuses to respond, you will get a refusal message
if (math_reasoning.refusal) {
console.log(math_reasoning.refusal);
} else {
console.log(math_reasoning.parsed);
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30class Step(BaseModel):
explanation: str
output: str
class MathReasoning(BaseModel):
steps: list[Step]
final_answer: str
completion = client.chat.completions.parse(
model="gpt-6-astra",
messages=[
{
"role": "system",
"content": "You are a helpful math tutor. Guide the user through the solution step by step.",
},
{"role": "user", "content": "how can I solve 8x + 7 = -23"},
],
response_format=MathReasoning,
)
math_reasoning = completion.choices[0].message
# If the model refuses to respond, you will get a refusal message
if math_reasoning.refusal:
print(math_reasoning.refusal)
else:
print(math_reasoning.parsed)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/shared"
)
func main() {
client := openai.NewClient()
completion, err := client.Chat.Completions.New(context.Background(), openai.ChatCompletionNewParams{
Model: "gpt-6-astra",
Messages: []openai.ChatCompletionMessageParamUnion{
openai.SystemMessage("You are a helpful math tutor. Guide the user through the solution step by step."),
openai.UserMessage("how can I solve 8x + 7 = -23"),
},
ResponseFormat: openai.ChatCompletionNewParamsResponseFormatUnion{
OfJSONSchema: &shared.ResponseFormatJSONSchemaParam{JSONSchema: shared.ResponseFormatJSONSchemaJSONSchemaParam{
Name: "math_reasoning", Schema: mathSchema(), Strict: openai.Bool(true),
}},
},
})
if err != nil {
panic(err)
}
message := completion.Choices[0].Message
if message.Refusal != "" {
fmt.Println(message.Refusal)
return
}
fmt.Println(message.Content)
}
func mathSchema() map[string]any {
return map[string]any{
"type": "object",
"properties": map[string]any{
"steps": map[string]any{"type": "array", "items": map[string]any{"type": "object", "properties": map[string]any{"explanation": map[string]any{"type": "string"}, "output": map[string]any{"type": "string"}}, "required": []string{"explanation", "output"}, "additionalProperties": false}},
"final_answer": map[string]any{"type": "string"},
},
"required": []string{"steps", "final_answer"},
"additionalProperties": false,
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.chat.completions.ChatCompletionCreateParams;
import java.util.List;
import java.util.Map;
Map<String, Object> schema =
Map.of(
"type",
"object",
"properties",
Map.of(
"steps",
Map.of(
"type",
"array",
"items",
Map.of(
"type",
"object",
"properties",
Map.of(
"explanation", Map.of("type", "string"),
"output", Map.of("type", "string")),
"required",
List.of("explanation", "output"),
"additionalProperties",
false)),
"final_answer", Map.of("type", "string")),
"required",
List.of("steps", "final_answer"),
"additionalProperties",
false);
ChatCompletionCreateParams params =
ChatCompletionCreateParams.builder()
.model("gpt-6-astra")
.addSystemMessage(
"You are a helpful math tutor. Guide the user through the solution step by step.")
.addUserMessage("How can I solve 8x + 7 = -23?")
.putAdditionalBodyProperty(
"response_format",
JsonValue.from(
Map.of(
"type",
"json_schema",
"json_schema",
Map.of("name", "math_reasoning", "strict", true, "schema", schema))))
.build();
var message = client.chat().completions().create(params).choices().get(0).message();
System.out.println(message.refusal().or(() -> message.content()).orElseThrow());1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51using OpenAI.Chat;
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
string model = "gpt-6-astra";
ChatClient client = new(model, key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": { "type": "string" },
"output": { "type": "string" }
},
"required": ["explanation", "output"],
"additionalProperties": false
}
},
"final_answer": { "type": "string" }
},
"required": ["steps", "final_answer"],
"additionalProperties": false
}
"""
);
ChatCompletionOptions options = new()
{
ResponseFormat = ChatResponseFormat.CreateJsonSchemaFormat(
"math_response",
schema,
jsonSchemaIsStrict: true
),
};
ChatCompletion completion = await client.CompleteChatAsync(
[new SystemChatMessage("You are a helpful math tutor. Guide the user through the solution step by step."), new UserChatMessage("How can I solve 8x + 7 = -23?")],
options
);
if (!string.IsNullOrEmpty(completion.Refusal))
{
Console.WriteLine(completion.Refusal);
}
else
{
Console.WriteLine(completion.Content[0].Text);
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48require "openai"
client = OpenAI::Client.new
math_schema = {
type: :object,
properties: {
steps: {
type: :array,
items: {
type: :object,
properties: {
explanation: { type: :string },
output: { type: :string }
},
required: %w[explanation output],
additionalProperties: false
}
},
final_answer: { type: :string }
},
required: %w[steps final_answer],
additionalProperties: false
}
completion = client.chat.completions.create(
model: "gpt-6-astra",
messages: [
{
role: :system,
content: "You are a helpful math tutor. Guide the user through the solution step by step."
},
{
role: :user,
content: "How can I solve 8x + 7 = -23?"
}
],
response_format: {
type: :json_schema,
json_schema: {
name: "math_reasoning",
strict: true,
schema: math_schema
}
}
)
message = completion.choices.fetch(0).message
puts(message.refusal || message.content)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44const Step = z.object({
explanation: z.string(),
output: z.string(),
});
const MathReasoning = z.object({
steps: z.array(Step),
final_answer: z.string(),
});
const response = await openai.responses.parse({
model: "gpt-6-astra",
input: [
{
role: "system",
content:
"You are a helpful math tutor. Guide the user through the solution step by step.",
},
{ role: "user", content: "how can I solve 8x + 7 = -23" },
],
text: {
format: zodTextFormat(MathReasoning, "math_response"),
},
});
for (const output of response.output) {
if (output.type !== "message") {
continue;
}
for (const item of output.content) {
if (item.type == "refusal") {
// If the model refuses to respond, you will get a refusal message
console.log(item.refusal);
continue;
}
if (!item.parsed) {
throw new Error("Could not parse response");
}
console.log(item.parsed);
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36class Step(BaseModel):
explanation: str
output: str
class MathReasoning(BaseModel):
steps: list[Step]
final_answer: str
response = client.responses.parse(
model="gpt-6-astra",
input=[
{
"role": "system",
"content": "You are a helpful math tutor. Guide the user through the solution step by step.",
},
{"role": "user", "content": "how can I solve 8x + 7 = -23"},
],
text_format=MathReasoning,
)
for output in response.output:
if output.type != "message":
continue
for item in output.content:
if item.type == "refusal":
# If the model refuses to respond, you will get a refusal message
print(item.refusal)
continue
if not item.parsed:
raise Exception("Could not parse response")
print(item.parsed)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57package main
import (
"context"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/responses"
)
func main() {
client := openai.NewClient()
response, err := client.Responses.New(context.Background(), responses.ResponseNewParams{
Model: "gpt-6-astra",
Input: responses.ResponseNewParamsInputUnion{OfInputItemList: responses.ResponseInputParam{
responses.ResponseInputItemParamOfMessage(
responses.ResponseInputMessageContentListParam{responses.ResponseInputContentParamOfInputText("You are a helpful math tutor. Guide the user through the solution step by step.")},
responses.EasyInputMessageRoleSystem,
),
responses.ResponseInputItemParamOfMessage(
responses.ResponseInputMessageContentListParam{responses.ResponseInputContentParamOfInputText("how can I solve 8x + 7 = -23")},
responses.EasyInputMessageRoleUser,
),
}},
Text: responses.ResponseTextConfigParam{Format: responses.ResponseFormatTextConfigUnionParam{
OfJSONSchema: &responses.ResponseFormatTextJSONSchemaConfigParam{Name: "math_response", Schema: mathSchema(), Strict: openai.Bool(true)},
}},
})
if err != nil {
panic(err)
}
for _, output := range response.Output {
if output.Type != "message" {
continue
}
for _, content := range output.AsMessage().Content {
if content.Type == "refusal" {
fmt.Println(content.AsRefusal().Refusal)
continue
}
fmt.Println(content.AsOutputText().Text)
}
}
}
func mathSchema() map[string]any {
return map[string]any{
"type": "object",
"properties": map[string]any{
"steps": map[string]any{"type": "array", "items": map[string]any{"type": "object", "properties": map[string]any{"explanation": map[string]any{"type": "string"}, "output": map[string]any{"type": "string"}}, "required": []string{"explanation", "output"}, "additionalProperties": false}},
"final_answer": map[string]any{"type": "string"},
},
"required": []string{"steps", "final_answer"},
"additionalProperties": false,
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.responses.EasyInputMessage;
import com.openai.models.responses.ResponseCreateParams;
import com.openai.models.responses.ResponseFormatTextJsonSchemaConfig;
import com.openai.models.responses.ResponseInputItem;
import com.openai.models.responses.ResponseTextConfig;
import java.util.List;
import java.util.Map;
Map<String, Object> schema =
Map.of(
"type",
"object",
"properties",
Map.of(
"steps",
Map.of(
"type",
"array",
"items",
Map.of(
"type",
"object",
"properties",
Map.of(
"explanation", Map.of("type", "string"),
"output", Map.of("type", "string")),
"required",
List.of("explanation", "output"),
"additionalProperties",
false)),
"final_answer", Map.of("type", "string")),
"required",
List.of("steps", "final_answer"),
"additionalProperties",
false);
ResponseCreateParams params =
ResponseCreateParams.builder()
.model("gpt-6-astra")
.inputOfResponse(
List.of(
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.SYSTEM)
.content(
"You are a helpful math tutor. Guide the user through the solution step by step.")
.build()),
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.USER)
.content("How can I solve 8x + 7 = -23?")
.build())))
.text(
ResponseTextConfig.builder()
.format(
ResponseFormatTextJsonSchemaConfig.builder()
.name("math_reasoning")
.strict(true)
.schema(
JsonValue.from(schema)
.convert(ResponseFormatTextJsonSchemaConfig.Schema.class))
.build())
.build())
.build();
var response = client.responses().create(params);
for (var output : response.output()) {
if (output.message().isEmpty()) continue;
for (var content : output.message().orElseThrow().content()) {
if (content.refusal().isPresent()) {
System.out.println(content.refusal().orElseThrow().refusal());
} else {
content.outputText().ifPresent(text -> System.out.println(text.text()));
}
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55using OpenAI.Responses;
#pragma warning disable OPENAI001
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ResponsesClient client = new(key);
BinaryData schema = BinaryData.FromString(
"""
{
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"type": "object",
"properties": {
"explanation": { "type": "string" },
"output": { "type": "string" }
},
"required": ["explanation", "output"],
"additionalProperties": false
}
},
"final_answer": { "type": "string" }
},
"required": ["steps", "final_answer"],
"additionalProperties": false
}
"""
);
CreateResponseOptions options = new()
{
Model = "gpt-6-astra",
TextOptions = new ResponseTextOptions
{
TextFormat = ResponseTextFormat.CreateJsonSchemaFormat(
"math_response",
schema,
jsonSchemaIsStrict: true
),
},
};
options.InputItems.Add(ResponseItem.CreateSystemMessageItem("You are a helpful math tutor. Guide the user through the solution step by step."));
options.InputItems.Add(ResponseItem.CreateUserMessageItem("How can I solve 8x + 7 = -23?"));
ResponseResult response = await client.CreateResponseAsync(options);
foreach (MessageResponseItem message in response.OutputItems.OfType<MessageResponseItem>())
{
foreach (ResponseContentPart content in message.Content)
{
Console.WriteLine(
content.Kind == ResponseContentPartKind.Refusal ? content.Refusal : content.Text
);
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58require "openai"
client = OpenAI::Client.new
math_schema = {
type: :object,
properties: {
steps: {
type: :array,
items: {
type: :object,
properties: {
explanation: { type: :string },
output: { type: :string }
},
required: %w[explanation output],
additionalProperties: false
}
},
final_answer: { type: :string }
},
required: %w[steps final_answer],
additionalProperties: false
}
response = client.responses.create(
model: "gpt-6-astra",
input: [
{
role: :system,
content: "You are a helpful math tutor. Guide the user through the solution step by step."
},
{
role: :user,
content: "How can I solve 8x + 7 = -23?"
}
],
text: {
format: {
type: :json_schema,
name: "math_response",
strict: true,
schema: math_schema
}
}
)
response.output.each do |item|
next unless item.is_a?(OpenAI::Models::Responses::ResponseOutputMessage)
item.content.each do |content|
case content
when OpenAI::Models::Responses::ResponseOutputRefusal
puts(content.refusal)
when OpenAI::Models::Responses::ResponseOutputText
puts(content.text)
end
end
end모델이 요청을 거부했을 때의 API 응답 예시는 다음과 같습니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28{
"id": "chatcmpl-9nYAG9LPNonX8DAyrkwYfemr3C8HC",
"object": "chat.completion",
"created": 1721596428,
"model": "gpt-4o-2024-08-06",
"choices": [
{
"index": 0,
"message": {
"role": "assistant",
"refusal": "I'm sorry, I cannot assist with that request."
},
"logprobs": null,
"finish_reason": "stop"
}
],
"usage": {
"prompt_tokens": 81,
"completion_tokens": 11,
"total_tokens": 92,
"completion_tokens_details": {
"reasoning_tokens": 0,
"accepted_prediction_tokens": 0,
"rejected_prediction_tokens": 0
}
},
"system_fingerprint": "fp_3407719c7f"
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32{
"id": "resp_1234567890",
"object": "response",
"created_at": 1721596428,
"status": "completed",
"completed_at": 1721596429,
"error": null,
"incomplete_details": null,
"input": [],
"instructions": null,
"max_output_tokens": null,
"model": "gpt-4o-2024-08-06",
"output": [{
"id": "msg_1234567890",
"type": "message",
"role": "assistant",
"content": [
{
"type": "refusal",
"refusal": "I'm sorry, I cannot assist with that request."
}
]
}],
"usage": {
"input_tokens": 81,
"output_tokens": 11,
"total_tokens": 92,
"output_tokens_details": {
"reasoning_tokens": 0,
}
},
}팁과 권장 사항
사용자 입력 처리하기
애플리케이션에서 사용자 입력을 받는다면 입력 내용으로 유효한 응답을 만들 수 없을 때 어떻게 처리해야 하는지 프롬프트에 명시하세요.
모델은 항상 지정한 스키마를 따르려고 하므로 입력이 스키마와 전혀 관련이 없으면 내용을 지어낼 수 있습니다.
예를 들어 모델이 입력 내용이 작업에 적합하지 않다고 판단하면 매개변수 값을 비워 두거나 특정 문장을 반환하도록 프롬프트에 지시할 수 있습니다.
오류 처리하기
구조화된 출력값에도 잘못된 내용이 포함될 수 있습니다. 오류가 발견되면 지시를 조정하거나, 시스템 지침에 예시를 추가하거나, 작업을 더 간단한 하위 작업으로 나눠 보세요. 입력을 개선하는 방법은 프롬프트 엔지니어링 가이드를 참고하세요.
JSON 스키마와 타입 간 불일치 방지하기
SDK에서 스키마 헬퍼를 제공한다면 이를 사용하는 것을 적극 권장합니다. JSON 스키마와 프로그래밍 언어에서 정의한 타입이 서로 어긋나는 것을 방지할 수 있습니다.
JSON 스키마를 직접 지정하려면 JSON 스키마나 해당 데이터 객체가 수정될 때 이를 알리는 CI 규칙을 추가할 수 있습니다. 타입 정의에서 JSON 스키마를 자동 생성하거나, 반대로 JSON 스키마에서 타입 정의를 자동 생성하는 CI 단계를 추가하는 방법도 있습니다.
스트리밍
스트리밍을 사용하면 모델 응답이나 함수 호출 인수를 생성되는 대로 처리하고 구조화된 데이터로 파싱할 수 있습니다.
이렇게 하면 응답이 완료될 때까지 기다리지 않고 처리를 시작할 수 있습니다. JSON 필드를 하나씩 표시하거나 함수 호출 인수가 준비되는 즉시 처리할 때 특히 유용합니다.
구조화된 출력값을 스트리밍할 때는 SDK를 사용하는 것이 좋습니다.
SDK의 stream 헬퍼 없이 함수 호출 인수를 스트리밍하는 예제는 함수 호출 가이드에서 확인할 수 있습니다.
다음 예제는 stream 헬퍼로 모델 응답을 스트리밍하는 방법을 보여 줍니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40import OpenAI from "openai";
import { zodResponseFormat } from "openai/helpers/zod";
import { z } from "zod";
const openai = new OpenAI();
const EntitiesSchema = z.object({
attributes: z.array(z.string()),
colors: z.array(z.string()),
animals: z.array(z.string()),
});
const stream = openai.chat.completions
.stream({
model: "gpt-6-astra",
messages: [
{ role: "system", content: "Extract entities from the input text" },
{
role: "user",
content:
"The quick brown fox jumps over the lazy dog with piercing blue eyes",
},
],
response_format: zodResponseFormat(EntitiesSchema, "entities"),
})
.on("refusal.done", () => console.log("request refused"))
.on("content.delta", ({ snapshot, parsed }) => {
console.log("content:", snapshot);
console.log("parsed:", parsed);
console.log();
})
.on("content.done", (props) => {
console.log(props);
});
await stream.done();
const finalCompletion = await stream.finalChatCompletion();
console.log(finalCompletion);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34from pydantic import BaseModel
from openai import OpenAI
class EntitiesModel(BaseModel):
attributes: list[str]
colors: list[str]
animals: list[str]
client = OpenAI()
with client.beta.chat.completions.stream(
model="gpt-6-astra",
messages=[
{"role": "system", "content": "Extract entities from the input text"},
{
"role": "user",
"content": "The quick brown fox jumps over the lazy dog with piercing blue eyes",
},
],
response_format=EntitiesModel,
) as stream:
for event in stream:
if event.type == "content.delta":
if event.parsed is not None: # Print the parsed data as JSON
print("content.delta parsed:", event.parsed)
elif event.type == "content.done":
print("content.done")
elif event.type == "error":
print("Error in stream:", event.error)
final_completion = stream.get_final_completion()
print("Final completion:", final_completion)stream 헬퍼로 함수 호출 인수도 파싱할 수 있습니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38import { zodFunction } from "openai/helpers/zod";
import OpenAI from "openai/index";
import { z } from "zod";
const GetWeatherArgs = z.object({
city: z.string(),
country: z.string(),
});
const client = new OpenAI();
const stream = client.chat.completions
.stream({
model: "gpt-5.6-terra",
reasoning_effort: "none",
parallel_tool_calls: true,
messages: [
{
role: "user",
content: "What's the weather like in SF and London?",
},
],
tools: [zodFunction({ name: "get_weather", parameters: GetWeatherArgs })],
})
.on("tool_calls.function.arguments.delta", (props) =>
console.log("tool_calls.function.arguments.delta", props)
)
.on("tool_calls.function.arguments.done", (props) =>
console.log("tool_calls.function.arguments.done", props)
)
.on("refusal.delta", ({ delta }) => {
process.stdout.write(delta);
})
.on("refusal.done", () => console.log("request refused"));
const completion = await stream.finalChatCompletion();
console.log("final completion:", completion);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34from pydantic import BaseModel
import openai
from openai import OpenAI
class GetWeather(BaseModel):
city: str
country: str
client = OpenAI()
with client.beta.chat.completions.stream(
model="gpt-5.6-terra",
reasoning_effort="none",
messages=[
{
"role": "user",
"content": "What's the weather like in SF and London?",
},
],
tools=[
openai.pydantic_function_tool(GetWeather, name="get_weather"),
],
parallel_tool_calls=True,
) as stream:
for event in stream:
if (
event.type == "tool_calls.function.arguments.delta"
or event.type == "tool_calls.function.arguments.done"
):
print(event)
print(stream.get_final_completion())1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37import { OpenAI } from "openai";
import { zodTextFormat } from "openai/helpers/zod";
import { z } from "zod";
const EntitiesSchema = z.object({
attributes: z.array(z.string()),
colors: z.array(z.string()),
animals: z.array(z.string()),
});
const openai = new OpenAI();
const stream = openai.responses
.stream({
model: "gpt-6-astra",
input: [
{ role: "user", content: "What's the weather like in Paris today?" },
],
text: {
format: zodTextFormat(EntitiesSchema, "entities"),
},
})
.on("response.refusal.delta", (event) => {
process.stdout.write(event.delta);
})
.on("response.output_text.delta", (event) => {
process.stdout.write(event.delta);
})
.on("response.output_text.done", () => {
process.stdout.write("\n");
})
.on("error", (error) => {
console.error(error);
});
const result = await stream.finalResponse();
console.log(result);1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35from openai import OpenAI
from pydantic import BaseModel
class EntitiesModel(BaseModel):
attributes: list[str]
colors: list[str]
animals: list[str]
client = OpenAI()
with client.responses.stream(
model="gpt-6-astra",
input=[
{"role": "system", "content": "Extract entities from the input text"},
{
"role": "user",
"content": "The quick brown fox jumps over the lazy dog with piercing blue eyes",
},
],
text_format=EntitiesModel,
) as stream:
for event in stream:
if event.type == "response.refusal.delta":
print(event.delta, end="")
elif event.type == "response.output_text.delta":
print(event.delta, end="")
elif event.type == "response.error":
print(event.error, end="")
elif event.type == "response.completed":
print("Completed") # print(event.response.output)
final_response = stream.get_final_response()
print(final_response)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.core.http.StreamResponse;
import com.openai.models.responses.EasyInputMessage;
import com.openai.models.responses.ResponseCreateParams;
import com.openai.models.responses.ResponseFormatTextJsonSchemaConfig;
import com.openai.models.responses.ResponseInputItem;
import com.openai.models.responses.ResponseStreamEvent;
import com.openai.models.responses.ResponseTextConfig;
import java.util.List;
import java.util.Map;
ResponseCreateParams params =
ResponseCreateParams.builder()
.model("gpt-6-astra")
.inputOfResponse(
List.of(
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.SYSTEM)
.content("Extract entities from the input text")
.build()),
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.USER)
.content(
"The quick brown fox jumps over the lazy dog with piercing blue eyes")
.build())))
.text(
ResponseTextConfig.builder()
.format(
ResponseFormatTextJsonSchemaConfig.builder()
.name("entities")
.strict(true)
.schema(
ResponseFormatTextJsonSchemaConfig.Schema.builder()
.putAdditionalProperty("type", JsonValue.from("object"))
.putAdditionalProperty(
"properties",
JsonValue.from(
Map.of(
"attributes",
Map.of(
"type",
"array",
"items",
Map.of("type", "string")),
"colors",
Map.of(
"type",
"array",
"items",
Map.of("type", "string")),
"animals",
Map.of(
"type",
"array",
"items",
Map.of("type", "string")))))
.putAdditionalProperty(
"required",
JsonValue.from(List.of("attributes", "colors", "animals")))
.putAdditionalProperty(
"additionalProperties", JsonValue.from(false))
.build())
.build())
.build())
.build();
try (StreamResponse<ResponseStreamEvent> stream = client.responses().createStreaming(params)) {
stream.stream()
.forEach(
event -> {
event.outputTextDelta().ifPresent(delta -> System.out.print(delta.delta()));
event.refusalDelta().ifPresent(refusal -> System.out.print(refusal.delta()));
event.error().ifPresent(error -> System.out.println(error.message()));
event
.completed()
.ifPresent(
completed -> {
System.out.println("Completed");
System.out.println(completed.response());
});
});
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56require "openai"
client = OpenAI::Client.new
entities_schema = {
type: :object,
properties: {
attributes: {
type: :array,
items: { type: :string }
},
colors: {
type: :array,
items: { type: :string }
},
animals: {
type: :array,
items: { type: :string }
}
},
required: %w[attributes colors animals],
additionalProperties: false
}
stream = client.responses.stream(
model: "gpt-6-astra",
input: [
{
role: :system,
content: "Extract entities from the input text."
},
{
role: :user,
content: "The quick brown fox jumps over the lazy dog with piercing blue eyes."
}
],
text: {
format: {
type: :json_schema,
name: "entities",
strict: true,
schema: entities_schema
}
}
)
stream.each do |event|
case event
when OpenAI::Models::Responses::ResponseRefusalDeltaEvent,
OpenAI::Models::Responses::ResponseTextDeltaEvent
print(event.delta)
when OpenAI::Models::Responses::ResponseErrorEvent
warn(event.message)
when OpenAI::Models::Responses::ResponseCompletedEvent
puts("\nCompleted")
end
end지원되는 스키마
구조화된 출력값은 JSON 스키마 문법의 일부를 지원합니다.
지원하는 타입
구조화된 출력값은 다음 타입을 지원합니다.
- 문자열
- 숫자
- 불리언
- 정수
- 객체
- 배열
- 열거형
- anyOf
지원하는 속성
속성의 타입을 지정할 때 다음 제약 조건도 함께 설정할 수 있습니다.
string에서 지원하는 속성
pattern— 문자열이 일치해야 하는 정규 표현식입니다.format— 미리 정의된 문자열 형식을 지정합니다. 현재 지원하는 형식은 다음과 같습니다.date-timetimedatedurationemailhostnameipv4ipv6uuid
number에서 지원하는 속성
multipleOf— 숫자는 지정한 값의 배수여야 합니다.maximum— 숫자는 지정한 값보다 작거나 같아야 합니다.exclusiveMaximum— 숫자는 지정한 값보다 작아야 합니다.minimum— 숫자는 지정한 값보다 크거나 같아야 합니다.exclusiveMinimum— 숫자는 지정한 값보다 커야 합니다.
array에서 지원하는 속성
minItems— 배열의 항목 수는 지정한 값 이상이어야 합니다.maxItems— 배열의 항목 수는 지정한 값 이하여야 합니다.
다음은 타입별 제약 조건을 사용하는 예시입니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27{
"name": "user_data",
"strict": true,
"schema": {
"type": "object",
"properties": {
"name": {
"type": "string",
"description": "The name of the user"
},
"username": {
"type": "string",
"description": "The username of the user. Must start with @",
"pattern": "^@[a-zA-Z0-9_]+$"
},
"email": {
"type": "string",
"description": "The email of the user",
"format": "email"
}
},
"additionalProperties": false,
"required": [
"name", "username", "email"
]
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28{
"name": "weather_data",
"strict": true,
"schema": {
"type": "object",
"properties": {
"location": {
"type": "string",
"description": "The location to get the weather for"
},
"unit": {
"type": ["string", "null"],
"description": "The unit to return the temperature in",
"enum": ["F", "C"]
},
"value": {
"type": "number",
"description": "The actual temperature value in the location",
"minimum": -130,
"maximum": 130
}
},
"additionalProperties": false,
"required": [
"location", "unit", "value"
]
}
}이러한 제약 조건은 파인튜닝된 모델에서 아직 지원되지 않습니다.
스키마의 루트는 객체여야 하며 anyOf를 사용할 수 없습니다
스키마의 최상위 요소는 객체여야 하며, 이 위치에서는 anyOf를 사용할 수 없습니다. 예를 들어 Zod에서 판별 유니온을 사용하면 최상위에 anyOf가 생성됩니다. 따라서 다음과 같은 코드는 작동하지 않습니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17import { z } from "zod";
import { zodResponseFormat } from "openai/helpers/zod";
const BaseResponseSchema = z.object({
/* ... */
});
const UnsuccessfulResponseSchema = z.object({
/* ... */
});
const finalSchema = z.discriminatedUnion("status", [
BaseResponseSchema,
UnsuccessfulResponseSchema,
]);
// Invalid JSON Schema for Structured Outputs
const json = zodResponseFormat(finalSchema, "final_schema");모든 필드는 required로 지정해야 합니다
구조화된 출력값을 사용하려면 모든 필드 또는 함수 매개변수를 required로 지정해야 합니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21{
"name": "get_weather",
"description": "Fetches the weather in the given location",
"strict": true,
"parameters": {
"type": "object",
"properties": {
"location": {
"type": "string",
"description": "The location to get the weather for"
},
"unit": {
"type": "string",
"description": "The unit to return the temperature in",
"enum": ["F", "C"]
}
},
"additionalProperties": false,
"required": ["location", "unit"]
}
}모든 필드는 필수이며 모델은 각 매개변수의 값을 반환합니다. 선택적 매개변수처럼 동작하게 하려면 null을 포함하는 유니온 타입을 사용할 수 있습니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23{
"name": "get_weather",
"description": "Fetches the weather in the given location",
"strict": true,
"parameters": {
"type": "object",
"properties": {
"location": {
"type": "string",
"description": "The location to get the weather for"
},
"unit": {
"type": ["string", "null"],
"description": "The unit to return the temperature in",
"enum": ["F", "C"]
}
},
"additionalProperties": false,
"required": [
"location", "unit"
]
}
}객체의 중첩 깊이 및 크기 제한
스키마 전체에서 객체 속성은 최대 5,000개까지, 중첩은 최대 10단계까지 허용됩니다.
문자열의 총길이 제한
스키마에서 모든 속성 이름, 정의 이름, enum 값, const 값의 문자열 길이 합계는 120,000자를 넘을 수 없습니다.
열거형 크기 제한
스키마의 모든 열거형 속성을 통틀어 열거형 값은 최대 1,000개까지 지정할 수 있습니다.
문자열 값을 갖는 열거형 속성 하나에 값이 250개를 초과하면 그 값들의 문자열 길이 합계는 15,000자를 넘을 수 없습니다.
객체에는 항상 additionalProperties: false를 설정해야 합니다
additionalProperties는 객체에 JSON 스키마에서 정의하지 않은 키와 값을 추가로 포함할 수 있는지 결정합니다.
구조화된 출력값은 지정된 키와 값만 생성하므로 이 기능을 사용하려면 개발자가 additionalProperties: false를 설정해야 합니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23{
"name": "get_weather",
"description": "Fetches the weather in the given location",
"strict": true,
"schema": {
"type": "object",
"properties": {
"location": {
"type": "string",
"description": "The location to get the weather for"
},
"unit": {
"type": "string",
"description": "The unit to return the temperature in",
"enum": ["F", "C"]
}
},
"additionalProperties": false,
"required": [
"location", "unit"
]
}
}키 순서
구조화된 출력값을 사용하면 결과의 키는 스키마에 지정된 순서대로 출력됩니다.
아직 지원하지 않는 타입별 키워드
- 스키마 조합:
allOf,not,dependentRequired,dependentSchemas,if,then,else
파인튜닝된 모델에서는 다음 키워드도 지원하지 않습니다.
- 문자열:
minLength,maxLength,pattern,format - 숫자:
minimum,maximum,multipleOf - 객체:
patternProperties - 배열:
minItems,maxItems
strict: true를 지정해 구조화된 출력값을 활성화한 상태에서 지원하지 않는 JSON 스키마로 API를 호출하면 오류가 반환됩니다.
anyOf의 각 하위 스키마는 지원 범위 내에서 유효한 JSON 스키마여야 합니다
다음은 지원되는 anyOf 스키마의 예시입니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56{
"type": "object",
"properties": {
"item": {
"anyOf": [
{
"type": "object",
"description": "The user object to insert into the database",
"properties": {
"name": {
"type": "string",
"description": "The name of the user"
},
"age": {
"type": "number",
"description": "The age of the user"
}
},
"additionalProperties": false,
"required": [
"name",
"age"
]
},
{
"type": "object",
"description": "The address object to insert into the database",
"properties": {
"number": {
"type": "string",
"description": "The number of the address. Eg. for 123 main st, this would be 123"
},
"street": {
"type": "string",
"description": "The street name. Eg. for 123 main st, this would be main st"
},
"city": {
"type": "string",
"description": "The city of the address"
}
},
"additionalProperties": false,
"required": [
"number",
"street",
"city"
]
}
]
}
},
"additionalProperties": false,
"required": [
"item"
]
}스키마 정의 지원
하위 스키마를 한 번 정의하면 스키마 곳곳에서 참조할 수 있습니다. 다음은 간단한 예시입니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37{
"type": "object",
"properties": {
"steps": {
"type": "array",
"items": {
"$ref": "#/$defs/step"
}
},
"final_answer": {
"type": "string"
}
},
"$defs": {
"step": {
"type": "object",
"properties": {
"explanation": {
"type": "string"
},
"output": {
"type": "string"
}
},
"required": [
"explanation",
"output"
],
"additionalProperties": false
}
},
"required": [
"steps",
"final_answer"
],
"additionalProperties": false
}재귀 스키마 지원
다음은 #로 루트를 다시 참조하는 재귀 스키마의 예시입니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47{
"name": "ui",
"description": "Dynamically generated UI",
"strict": true,
"schema": {
"type": "object",
"properties": {
"type": {
"type": "string",
"description": "The type of the UI component",
"enum": ["div", "button", "header", "section", "field", "form"]
},
"label": {
"type": "string",
"description": "The label of the UI component, used for buttons or form fields"
},
"children": {
"type": "array",
"description": "Nested UI components",
"items": {
"$ref": "#"
}
},
"attributes": {
"type": "array",
"description": "Arbitrary attributes for the UI component, suitable for any element",
"items": {
"type": "object",
"properties": {
"name": {
"type": "string",
"description": "The name of the attribute, for example onClick or className"
},
"value": {
"type": "string",
"description": "The value of the attribute"
}
},
"additionalProperties": false,
"required": ["name", "value"]
}
}
},
"required": ["type", "label", "children", "attributes"],
"additionalProperties": false
}
}다음은 재귀 참조를 명시적으로 지정한 스키마의 예시입니다.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37{
"type": "object",
"properties": {
"linked_list": {
"$ref": "#/$defs/linked_list_node"
}
},
"$defs": {
"linked_list_node": {
"type": "object",
"properties": {
"value": {
"type": "number"
},
"next": {
"anyOf": [
{
"$ref": "#/$defs/linked_list_node"
},
{
"type": "null"
}
]
}
},
"additionalProperties": false,
"required": [
"next",
"value"
]
}
},
"additionalProperties": false,
"required": [
"linked_list"
]
}JSON 모드
JSON 모드는 구조화된 출력값의 기본적인 형태입니다. JSON 모드는 모델이 유효한 JSON을 출력하도록 보장하며, 구조화된 출력값은 모델 출력이 지정한 스키마까지 따르도록 보장합니다. 사용하려는 환경에서 구조화된 출력값을 지원한다면 이 기능을 사용하는 것이 좋습니다.
JSON 모드를 활성화하면 일부 예외적인 경우를 제외하고 모델이 유효한 JSON을 출력합니다. 이러한 예외는 애플리케이션에서 감지해 적절히 처리해야 합니다.
Chat Completions API에서 JSON 모드를 활성화하려면 response_format을 { "type": "json_object" }로 설정하세요. 함수 호출을 사용하면 JSON 모드가 항상 활성화됩니다.
Responses API에서 JSON 모드를 활성화하려면 text.format을 { "type": "json_object" }로 설정하세요. 함수 호출을 사용하면 JSON 모드가 항상 활성화됩니다.
중요한 유의 사항은 다음과 같습니다.
- JSON 모드를 사용할 때는 시스템 메시지 등 대화에 포함된 메시지에서 모델에 JSON을 생성하라고 반드시 지시해야 합니다. 명시적으로 지시하지 않으면 모델이 공백 문자를 계속 생성해 토큰 한도에 도달할 때까지 요청이 끝나지 않을 수 있습니다. 이 지시가 누락되는 것을 방지하기 위해 API는 컨텍스트 어디에도 “JSON”이라는 문자열이 없으면 오류를 반환합니다.
- JSON 모드는 출력이 유효한 JSON이고 오류 없이 파싱된다는 점만 보장합니다. 특정 스키마를 따르도록 보장하려면 구조화된 출력값을 사용하세요. 사용할 수 없다면 검증 라이브러리로 출력이 원하는 스키마를 따르는지 확인하고 필요에 따라 요청을 다시 보내세요.
- 모델이 완전한 JSON 객체를 출력하지 못하는 예외 상황을 애플리케이션에서 감지하고 처리해야 합니다(아래 참고).
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52const we_did_not_specify_stop_tokens = true;
try {
const response = await openai.chat.completions.create({
model: "gpt-6-astra",
messages: [
{
role: "system",
content: "You are a helpful assistant designed to output JSON.",
},
{
role: "user",
content:
"Who won the world series in 2020? Please respond in the format {winner: ...}",
},
],
store: true,
response_format: { type: "json_object" },
});
// Check if the conversation was too long for the context window, resulting in incomplete JSON
if (response.choices[0].finish_reason === "length") {
// your code should handle this error case
}
// Check if the OpenAI safety system refused the request and generated a refusal instead
if (response.choices[0].message.refusal) {
// your code should handle this error case
// In this case, the .content field will contain the explanation (if any) that the model generated for why it is refusing
console.log(response.choices[0].message.refusal);
}
// Check if the model's output included restricted content, so the generation of JSON was halted and may be partial
if (response.choices[0].finish_reason === "content_filter") {
// your code should handle this error case
}
if (response.choices[0].finish_reason === "stop") {
// In this case the model has either successfully finished generating the JSON object according to your schema, or the model generated one of the tokens you provided as a "stop token"
if (we_did_not_specify_stop_tokens) {
// If you didn't specify any stop tokens, then the generation is complete and the content key will contain the serialized JSON object
// This will parse successfully and should now contain {"winner": "Los Angeles Dodgers"}
console.log(JSON.parse(response.choices[0].message.content));
} else {
// Check if the response.choices[0].message.content ends with one of your stop tokens and handle appropriately
}
}
} catch (e) {
// Your code should handle errors here, for example a network error calling the API
console.error(e);
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42we_did_not_specify_stop_tokens = True
try:
response = client.chat.completions.create(
model="gpt-6-astra",
messages=[
{
"role": "system",
"content": "You are a helpful assistant designed to output JSON.",
},
{
"role": "user",
"content": 'Who won the World Series in 2020? Respond as {"winner": "team name"}.',
},
],
response_format={"type": "json_object"},
)
# Check if the conversation was too long for the context window, resulting in incomplete JSON
if response.choices[0].finish_reason == "length":
raise RuntimeError("The response was truncated before the JSON completed.")
# Check if the OpenAI safety system refused the request and generated a refusal instead
if response.choices[0].message.refusal:
# your code should handle this error case
# In this case, the .content field will contain the explanation (if any) that the model generated for why it is refusing
print(response.choices[0].message.refusal)
# Check if the model's output included restricted content, so the generation of JSON was halted and may be partial
if response.choices[0].finish_reason == "content_filter":
raise RuntimeError("The response was interrupted by the content filter.")
if response.choices[0].finish_reason == "stop":
# In this case the model has either successfully finished generating the JSON object according to your schema, or the model generated one of the tokens you provided as a "stop token"
if we_did_not_specify_stop_tokens:
# If you didn't specify any stop tokens, then the generation is complete and the content key will contain the serialized JSON object
# This will parse successfully and should now contain "{"winner": "Los Angeles Dodgers"}"
print(response.choices[0].message.content)
except Exception as e:
# Your code should handle errors here, for example a network error calling the API
print(e)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44package main
import (
"context"
"encoding/json"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/shared"
)
func main() {
client := openai.NewClient()
completion, err := client.Chat.Completions.New(context.Background(), openai.ChatCompletionNewParams{
Model: "gpt-6-astra",
Messages: []openai.ChatCompletionMessageParamUnion{
openai.SystemMessage("You are a helpful assistant designed to output JSON."),
openai.UserMessage("Who won the world series in 2020? Please respond in the format {winner: ...}"),
},
ResponseFormat: openai.ChatCompletionNewParamsResponseFormatUnion{
OfJSONObject: &shared.ResponseFormatJSONObjectParam{},
},
})
if err != nil {
panic(err)
}
choice := completion.Choices[0]
if choice.FinishReason == "length" || choice.FinishReason == "content_filter" {
fmt.Println("The JSON response is incomplete.")
return
}
if choice.Message.Refusal != "" {
fmt.Println(choice.Message.Refusal)
return
}
if choice.FinishReason == "stop" {
var value map[string]any
if err := json.Unmarshal([]byte(choice.Message.Content), &value); err != nil {
panic(err)
}
fmt.Println(value)
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31import com.fasterxml.jackson.databind.ObjectMapper;
import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.chat.completions.ChatCompletion;
import com.openai.models.chat.completions.ChatCompletionCreateParams;
import java.io.IOException;
import java.util.Map;
ChatCompletionCreateParams params =
ChatCompletionCreateParams.builder()
.model("gpt-6-astra")
.addSystemMessage("You are a helpful assistant designed to output JSON.")
.addUserMessage("Who won the World Series in 2020? Respond as {winner: ...}.")
.putAdditionalBodyProperty(
"response_format", JsonValue.from(Map.of("type", "json_object")))
.build();
var choice = client.chat().completions().create(params).choices().get(0);
if (choice.finishReason().equals(ChatCompletion.Choice.FinishReason.LENGTH)
|| choice.finishReason().equals(ChatCompletion.Choice.FinishReason.CONTENT_FILTER)) {
System.out.println("The JSON response is incomplete.");
} else if (choice.message().refusal().isPresent()) {
System.out.println(choice.message().refusal().orElseThrow());
} else if (choice.finishReason().equals(ChatCompletion.Choice.FinishReason.STOP)) {
String content = choice.message().content().orElseThrow();
System.out.println(
new ObjectMapper()
.writerWithDefaultPrettyPrinter()
.writeValueAsString(new ObjectMapper().readTree(content)));
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35using OpenAI.Chat;
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
string model = "gpt-6-astra";
ChatClient client = new(model, key);
ChatCompletionOptions options = new()
{
ResponseFormat = ChatResponseFormat.CreateJsonObjectFormat(),
};
ChatCompletion completion = await client.CompleteChatAsync(
[new SystemChatMessage("You are a helpful assistant designed to output JSON."), new UserChatMessage("Who won the World Series in 2020? Respond with the winner in JSON.")],
options
);
if (completion.FinishReason == ChatFinishReason.Length)
{
Console.WriteLine("The response was truncated before the JSON completed.");
}
else if (completion.FinishReason == ChatFinishReason.ContentFilter)
{
Console.WriteLine("The response was interrupted by the content filter.");
}
else if (!string.IsNullOrEmpty(completion.Refusal))
{
Console.WriteLine(completion.Refusal);
}
else if (completion.FinishReason == ChatFinishReason.Stop && completion.Content.Count > 0)
{
Console.WriteLine(completion.Content[0].Text);
}
else
{
throw new InvalidOperationException("The completion did not contain a JSON response.");
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32require "json"
require "openai"
client = OpenAI::Client.new
completion = client.chat.completions.create(
model: "gpt-6-astra",
messages: [
{
role: :system,
content: "You are a helpful assistant designed to output JSON."
},
{
role: :user,
content: "Who won the World Series in 2020? Respond in the format {winner: ...}."
}
],
response_format: { type: :json_object }
)
choice = completion.choices.fetch(0)
finish_reason = choice.finish_reason
if [
OpenAI::Chat::ChatCompletion::Choice::FinishReason::LENGTH,
OpenAI::Chat::ChatCompletion::Choice::FinishReason::CONTENT_FILTER
].include?(finish_reason)
warn("The JSON response is incomplete.")
elsif choice.message.refusal
puts(choice.message.refusal)
elsif finish_reason == OpenAI::Chat::ChatCompletion::Choice::FinishReason::STOP
content = choice.message.content or raise "No response content"
puts(JSON.pretty_generate(JSON.parse(content)))
end1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60const we_did_not_specify_stop_tokens = true;
try {
const response = await openai.responses.create({
model: "gpt-6-astra",
input: [
{
role: "system",
content: "You are a helpful assistant designed to output JSON.",
},
{
role: "user",
content:
"Who won the world series in 2020? Please respond in the format {winner: ...}",
},
],
text: { format: { type: "json_object" } },
});
const message = response.output.find((item) => item.type === "message");
const messageContent = message?.content[0];
// Check if the conversation was too long for the context window, resulting in incomplete JSON
if (
response.status === "incomplete" &&
response.incomplete_details.reason === "max_output_tokens"
) {
// your code should handle this error case
}
// Check if the OpenAI safety system refused the request and generated a refusal instead
if (messageContent?.type === "refusal") {
// your code should handle this error case
// In this case, the .content field will contain the explanation (if any) that the model generated for why it is refusing
console.log(messageContent.refusal);
}
// Check if the model's output included restricted content, so the generation of JSON was halted and may be partial
if (
response.status === "incomplete" &&
response.incomplete_details.reason === "content_filter"
) {
// your code should handle this error case
}
if (response.status === "completed") {
// In this case the model has either successfully finished generating the JSON object according to your schema, or the model generated one of the tokens you provided as a "stop token"
if (we_did_not_specify_stop_tokens) {
// If you didn't specify any stop tokens, then the generation is complete and the content key will contain the serialized JSON object
// This will parse successfully and should now contain {"winner": "Los Angeles Dodgers"}
console.log(JSON.parse(response.output_text));
} else {
// Check if the response.output_text ends with one of your stop tokens and handle appropriately
}
}
} catch (e) {
// Your code should handle errors here, for example a network error calling the API
console.error(e);
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51we_did_not_specify_stop_tokens = True
try:
response = client.responses.create(
model="gpt-6-astra",
input=[
{
"role": "system",
"content": "You are a helpful assistant designed to output JSON.",
},
{
"role": "user",
"content": 'Who won the World Series in 2020? Respond as {"winner": "team name"}.',
},
],
text={"format": {"type": "json_object"}},
)
message = next((item for item in response.output if item.type == "message"), None)
message_content = message.content[0] if message and message.content else None
# Check if the conversation was too long for the context window, resulting in incomplete JSON
if (
response.status == "incomplete"
and response.incomplete_details.reason == "max_output_tokens"
):
raise RuntimeError("The response was truncated before the JSON completed.")
# Check if the OpenAI safety system refused the request and generated a refusal instead
if message_content and message_content.type == "refusal":
# your code should handle this error case
# In this case, the .content field will contain the explanation (if any) that the model generated for why it is refusing
print(message_content.refusal)
# Check if the model's output included restricted content, so the generation of JSON was halted and may be partial
if (
response.status == "incomplete"
and response.incomplete_details.reason == "content_filter"
):
raise RuntimeError("The response was interrupted by the content filter.")
if response.status == "completed":
# In this case the model has either successfully finished generating the JSON object according to your schema, or the model generated one of the tokens you provided as a "stop token"
if we_did_not_specify_stop_tokens:
# If you didn't specify any stop tokens, then the generation is complete and the content key will contain the serialized JSON object
# This will parse successfully and should now contain "{"winner": "Los Angeles Dodgers"}"
print(response.output_text)
except Exception as e:
# Your code should handle errors here, for example a network error calling the API
print(e)1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57package main
import (
"context"
"encoding/json"
"fmt"
"github.com/openai/openai-go/v3"
"github.com/openai/openai-go/v3/responses"
"github.com/openai/openai-go/v3/shared"
)
func main() {
client := openai.NewClient()
response, err := client.Responses.New(context.Background(), responses.ResponseNewParams{
Model: "gpt-6-astra",
Input: responses.ResponseNewParamsInputUnion{OfInputItemList: responses.ResponseInputParam{
responses.ResponseInputItemParamOfMessage(
responses.ResponseInputMessageContentListParam{responses.ResponseInputContentParamOfInputText("You are a helpful assistant designed to output JSON.")},
responses.EasyInputMessageRoleSystem,
),
responses.ResponseInputItemParamOfMessage(
responses.ResponseInputMessageContentListParam{responses.ResponseInputContentParamOfInputText("Who won the world series in 2020? Please respond in the format {winner: ...}")},
responses.EasyInputMessageRoleUser,
),
}},
Text: responses.ResponseTextConfigParam{Format: responses.ResponseFormatTextConfigUnionParam{
OfJSONObject: &shared.ResponseFormatJSONObjectParam{},
}},
})
if err != nil {
panic(err)
}
if response.Status == "incomplete" {
fmt.Println("The JSON response is incomplete.")
return
}
for _, output := range response.Output {
if output.Type != "message" {
continue
}
for _, content := range output.AsMessage().Content {
if content.Type == "refusal" {
fmt.Println(content.AsRefusal().Refusal)
return
}
}
}
if response.Status == "completed" {
var value map[string]any
if err := json.Unmarshal([]byte(response.OutputText()), &value); err != nil {
panic(err)
}
fmt.Println(value)
}
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.errors.OpenAIServiceException;
import com.openai.models.ResponseFormatJsonObject;
import com.openai.models.responses.EasyInputMessage;
import com.openai.models.responses.ResponseCreateParams;
import com.openai.models.responses.ResponseInputItem;
import com.openai.models.responses.ResponseStatus;
import com.openai.models.responses.ResponseTextConfig;
import java.util.List;
ResponseCreateParams params =
ResponseCreateParams.builder()
.model("gpt-6-astra")
.inputOfResponse(
List.of(
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.SYSTEM)
.content("You are a helpful assistant designed to output JSON.")
.build()),
ResponseInputItem.ofEasyInputMessage(
EasyInputMessage.builder()
.role(EasyInputMessage.Role.USER)
.content(
"Who won the World Series in 2020? Respond in the format {winner: ...}.")
.build())))
.text(
ResponseTextConfig.builder()
.format(ResponseFormatJsonObject.builder().build())
.build())
.build();
try {
var response = client.responses().create(params);
if (response.status().filter(ResponseStatus.INCOMPLETE::equals).isPresent()) {
String reason =
response
.incompleteDetails()
.flatMap(details -> details.reason())
.map(Object::toString)
.orElse("unknown");
System.out.println("The JSON response is incomplete. Reason: " + reason);
return;
}
for (var output : response.output()) {
if (output.message().isEmpty()) continue;
for (var content : output.message().orElseThrow().content()) {
if (content.refusal().isPresent()) {
System.out.println(content.refusal().orElseThrow().refusal());
return;
}
if (response.status().filter(ResponseStatus.COMPLETED::equals).isPresent()) {
content.outputText().ifPresent(text -> System.out.println(text.text()));
}
}
}
} catch (OpenAIServiceException error) {
System.out.println("Request failed: " + error.getMessage());
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46using OpenAI.Responses;
#pragma warning disable OPENAI001
string key = Environment.GetEnvironmentVariable("OPENAI_API_KEY")!;
ResponsesClient client = new(key);
CreateResponseOptions options = new()
{
Model = "gpt-6-astra",
TextOptions = new ResponseTextOptions
{
TextFormat = ResponseTextFormat.CreateJsonObjectFormat(),
},
};
options.InputItems.Add(ResponseItem.CreateSystemMessageItem("You are a helpful assistant designed to output JSON."));
options.InputItems.Add(ResponseItem.CreateUserMessageItem("Who won the World Series in 2020? Respond with the winner in JSON."));
ResponseResult response = await client.CreateResponseAsync(options);
if (
response.Status == ResponseStatus.Incomplete
&& response.IncompleteStatusDetails?.Reason == ResponseIncompleteStatusReason.MaxOutputTokens
)
{
Console.WriteLine("The response was truncated before the JSON completed.");
}
else if (
response.Status == ResponseStatus.Incomplete
&& response.IncompleteStatusDetails?.Reason == ResponseIncompleteStatusReason.ContentFilter
)
{
Console.WriteLine("The response was interrupted by the content filter.");
}
else if (response.Status == ResponseStatus.Completed)
{
MessageResponseItem message = response.OutputItems.OfType<MessageResponseItem>().FirstOrDefault()
?? throw new InvalidOperationException("The response did not include an output message.");
ResponseContentPart content = message.Content.FirstOrDefault()
?? throw new InvalidOperationException("The response did not include output content.");
Console.WriteLine(
content.Kind == ResponseContentPartKind.Refusal ? content.Refusal : content.Text
);
}
else
{
throw new InvalidOperationException($"The response ended with status: {response.Status}");
}1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33require "json"
require "openai"
client = OpenAI::Client.new
response = client.responses.create(
model: "gpt-6-astra",
input: [
{
role: :system,
content: "You are a helpful assistant designed to output JSON."
},
{
role: :user,
content: "Who won the World Series in 2020? Respond in the format {winner: ...}."
}
],
text: { format: { type: :json_object } }
)
if response.status == OpenAI::Responses::ResponseStatus::INCOMPLETE
warn("The JSON response is incomplete.")
else
refusal = response.output
.grep(OpenAI::Models::Responses::ResponseOutputMessage)
.flat_map(&:content)
.find { |content| content.is_a?(OpenAI::Models::Responses::ResponseOutputRefusal) }
if refusal.is_a?(OpenAI::Models::Responses::ResponseOutputRefusal)
puts(refusal.refusal)
elsif response.status == OpenAI::Responses::ResponseStatus::COMPLETED
puts(JSON.pretty_generate(JSON.parse(response.output_text)))
end
end참고 자료
구조화된 출력값에 대해 더 알아보려면 다음 자료를 참고하세요.
- 구조화된 출력값 입문 쿡북을 살펴보세요.
- 구조화된 출력값으로 멀티 에이전트 시스템을 구축하는 방법을 알아보세요.