JSON 是全球應用程式交換資料時最常用的格式之一。
模型未拒絕請求且已完整產生回應時,結構化輸出可確保該回應符合你提供的 JSON 結構描述。應用程式仍須處理拒絕回應、不完整的回應及 API 錯誤,並確認輸出符合結構描述未涵蓋的所有業務規則。
結構化輸出的優點包括:
- 符合結構描述: 只要你提供的是支援的 JSON 結構描述,輸出就會包含所有必要欄位,列舉值也會符合結構描述的規定。
- 明確標示拒絕回應: 現在可以透過程式偵測模型是否基於安全考量拒絕請求。
- 提示詞更簡單: 無須在提示詞中反覆強調格式要求,就能取得格式一致的回應。
除了 REST API 支援 JSON 結構描述,OpenAI 的 Python 和 JavaScript 程式庫也分別支援使用 pydantic.BaseModel 和 z.object 定義物件的結構描述。下列範例示範如何從非結構化文字中擷取資訊,讓結果符合程式碼中定義的結構描述。
Ruby SDK 支援以 Sorbet T::Struct 定義的結構描述,並會傳回具有型別的解析結果。
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(", "))支援的模型
從 GPT-4o 開始,我們的最新大型語言模型都支援結構化輸出。新專案建議從 gpt-6-astra 開始。gpt-4-turbo 及更早的模型則可改用 JSON 模式。
使用結構化輸出時,何時該用函式呼叫,何時該用 response_format
使用結構化輸出時,何時該用函式呼叫,何時該用 text.format
OpenAI API 提供兩種使用結構化輸出的方式:
- 使用函式呼叫
- 使用
json_schema回應格式
如果你要讓模型使用應用程式提供的功能,函式呼叫就很適合。
舉例來說,你可以讓模型呼叫查詢資料庫的函式,藉此打造能協助使用者處理訂單的 AI 助理,也可以讓模型呼叫能與使用者介面互動的函式。
如果你要指定的是模型回覆使用者時應遵循的結構描述,而非模型呼叫工具時使用的結構描述,則較適合透過 response_format 使用結構化輸出。
例如,開發數學輔導應用程式時,你可以讓助理依照特定的 JSON 結構描述回覆使用者,這樣就能產生使用者介面,以不同方式呈現模型輸出的各個部分。
實際使用時:
- 如果要讓模型使用系統中的工具、函式或資料等資源,請使用函式呼叫。如果要指定模型回覆使用者時的輸出結構,請使用結構化的
response_format。
- 如果要讓模型使用系統中的工具、函式或資料等資源,請使用函式呼叫。如果要指定模型回覆使用者時的輸出結構,請使用結構化的
text.format。
結構化輸出與 JSON 模式的差異
結構化輸出是 JSON 模式的進階版本。兩者都能確保產生有效的 JSON,但只有結構化輸出能確保結果符合指定的結構描述。Responses API、Chat Completions API、Assistants API、Fine-tuning API 和 Batch API 都支援結構化輸出與 JSON 模式。
只要可行,建議一律優先使用結構化輸出。
不過,透過 response_format: {type: "json_schema", ...} 使用結構化輸出時,只有 gpt-4o-mini、gpt-4o-mini-2024-07-18、gpt-4o-2024-08-06 及後續模型快照支援這項設定。
| 結構化輸出 | JSON 模式 | |
|---|---|---|
| 輸出有效的 JSON | 是 | 是 |
| 符合結構描述 | 是(請參閱支援的結構描述) | 否 |
| 相容模型 | gpt-4o-mini、gpt-4o-2024-08-06 及後續模型 |
gpt-3.5-turbo、gpt-4-*、gpt-4o-* 及相容的 GPT-5 模型 |
| 啟用方式 | response_format: { type: "json_schema", json_schema: {"strict": true, "schema": ...} } |
response_format: { type: "json_object" } |
| 結構化輸出 | JSON 模式 | |
|---|---|---|
| 輸出有效的 JSON | 是 | 是 |
| 符合結構描述 | 是(請參閱支援的結構描述) | 否 |
| 相容模型 | gpt-4o-mini、gpt-4o-2024-08-06 及後續模型 |
gpt-3.5-turbo、gpt-4-*、gpt-4o-* 及相容的 GPT-5 模型 |
| 啟用方式 | text: { format: { type: "json_schema", "strict": true, "schema": ... } } |
text: { format: { type: "json_object" } } |
範例
思路鏈
你可以要求模型以結構化格式逐步作答,引導使用者理解每個解題步驟。
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
}
}
}'回應範例
{
"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"
}擷取結構化資料
你可以定義結構化欄位,從研究論文等非結構化輸入資料中擷取各欄位的值。
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
}
}
}'回應範例
{
"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"
]
}產生使用者介面
你可以用遞迴資料結構來表示 HTML,並透過列舉等方式限制可用的值,藉此產生有效的 HTML。
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
48import OpenAI from "openai";
import { z } from "zod";
import { zodResponseFormat } from "openai/helpers/zod";
const openai = new OpenAI();
type UIElement = {
type: "div" | "button" | "header" | "section" | "field" | "form";
label: string;
children: UIElement[];
attributes: { name: string; value: string }[];
};
const UI: z.ZodType<UIElement> = 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",
max_completion_tokens: 4096,
messages: [
{
role: "system",
content: "You are a UI generator AI. Convert the user input into a UI.",
},
{
role: "user",
content:
"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.",
},
],
response_format: zodResponseFormat(UI, "ui"),
});
const ui = completion.choices[0].message.parsed;
if (!ui) throw new Error("UI generation did not complete.");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
38
39
40
41
42
43
44
45
46
47
48
49
50
51import OpenAI from "openai";
import { zodTextFormat } from "openai/helpers/zod";
import { z } from "zod";
const openai = new OpenAI();
type UIElement = {
type: "div" | "button" | "header" | "section" | "field" | "form";
label: string;
children: UIElement[];
attributes: { name: string; value: string }[];
};
const UI: z.ZodType<UIElement> = 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",
max_output_tokens: 4096,
input: [
{
role: "system",
content: "You are a UI generator AI. Convert the user input into a UI.",
},
{
role: "user",
content:
"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.",
},
],
text: {
format: zodTextFormat(UI, "ui"),
},
});
const ui = response.output_parsed;
if (response.status !== "completed" || !ui)
throw new Error("UI generation did not complete.");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
}
}
}'回應範例
{
"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"
}
]
}內容審核
你可以依據多個類別對輸入內容進行分類,這是內容審核的常見做法。
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
}
}
}'回應範例
{
"is_violating": false,
"category": null,
"explanation_if_violating": null
}如何搭配下列設定使用結構化輸出:response_format
使用結構化輸出時,你可以透過新的 SDK 輔助工具,將模型輸出解析為所需格式,也可以直接指定 JSON 結構描述。
注意: 使用微調模型時,首次以某個結構描述發出請求,API 需要先處理該結構描述,因此回應時間會較長。後續使用相同結構描述的請求不會有這項額外延遲。其他模型沒有這項限制。
首先,你必須定義一個物件或資料結構,用來表示模型輸出必須符合的 JSON 結構描述。可參考本指南開頭的範例。
結構化輸出支援 JSON 結構描述的大部分功能,但基於效能或技術上的考量,部分功能無法使用。詳情請參閱支援的結構描述。
例如,你可以用以下方式定義物件:
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: str資料結構設計建議
為了讓模型產生高品質的輸出,建議採取以下做法:
- 為鍵取清楚、容易理解的名稱
- 為資料結構中的重要鍵撰寫清楚的標題和說明
- 設計並執行評估,找出最適合你使用情境的資料結構
你可以使用 parse 方法,自動將 JSON 回應解析為你定義的物件。
SDK 會自動提供與資料結構對應的 JSON 結構描述,再將回應解析為物件。
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,
)在某些情況下,模型可能無法產生符合指定 JSON 結構描述的有效回應。
例如,模型可能基於安全考量而拒絕回答,也可能因輸出達到 Token 數量上限而無法產生完整的回應。
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)
end若要使用結構化輸出,只需指定下列參數:
response_format: { "type": "json_schema", "json_schema": … , "strict": true } text: { format: { type: "json_schema", "strict": true, "schema": … } } 例如:
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
}
}
}'注意: 首次使用某個結構描述發出請求時,API 需要處理該結構描述,因此會有額外延遲。後續使用相同結構描述的請求則不會有這項額外延遲。
在某些情況下,模型可能無法產生符合所提供的 JSON 結構描述的有效回應。
例如,模型可能基於安全考量而拒絕回答,或在達到 Token 數量上限時停止生成,導致回應不完整。
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)
end確認回應中包含符合結構描述的 JSON 後,請將該 JSON 解析為所用程式語言的原生資料結構。如果使用具備型別系統的語言,也可以用對應的型別或類別來表示這些資料。
例如:
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)如何搭配下列設定使用結構化輸出:text.format
若要使用結構化輸出,只需指定下列參數:
response_format: { "type": "json_schema", "json_schema": … , "strict": true } text: { format: { type: "json_schema", "strict": true, "schema": … } } 例如:
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
}
}
}'注意: 首次使用某個結構描述發出請求時,API 需要處理該結構描述,因此會有額外延遲。後續使用相同結構描述的請求則不會有這項額外延遲。
在某些情況下,模型可能無法產生符合所提供的 JSON 結構描述的有效回應。
例如,模型可能基於安全考量而拒絕回答,或在達到 Token 數量上限時停止生成,導致回應不完整。
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)
end確認回應中包含符合結構描述的 JSON 後,請將該 JSON 解析為所用程式語言的原生資料結構。如果使用具備型別系統的語言,也可以用對應的型別或類別來表示這些資料。
例如:
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)在對話中變更回應格式
若要讓下一次回應改用新的結構描述,請先將 configuration_update 項目附加到 input,再附加下一則使用者訊息。每次發出請求時,都要保留原本的頂層 text.format 設定,讓先前的提示詞前綴仍符合提示詞快取的條件。
在支援這項功能的 GPT-6 及後續模型上,使用標準單一智慧體模式時,可以更新回應格式。
假設上一輪對話以純文字說明某批貨物包含 17 個紅色貨箱和 26 個藍色貨箱。請使用 previous_response_id 延續對話,並將下列陣列作為 input 傳入。
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
end預期的答案是 {"total_crates":43}。回應物件中的 text.format 仍會顯示原本在請求層級指定的格式。先確認模型沒有拒絕請求,且回應完整,再使用最近一次格式更新所指定的結構描述解析答案。
每次更新格式都會取代整個格式設定。更新後的格式會持續套用到後續各輪對話,即使你只更新 reasoning.effort 也一樣。若要切換回純文字,請附加 {"type":"configuration_update","text":{"format":{"type":"text"}}}。格式更新僅接受 json_schema 和 text。
如果你手動重播對話記錄,請將每個更新項目保留在原來的位置。明確執行壓縮後,請重新指定你變更過的每一項設定。詳見設定更新的相容性。
使用結構化輸出時的拒絕回應
使用結構化輸出處理使用者提供的輸入時,OpenAI 模型偶爾會基於安全考量拒絕執行請求。由於拒絕回應不一定符合你在 response_format 中提供的結構描述,API 回應會包含名為 refusal 的新欄位,表示模型已拒絕執行請求。
如果輸出物件中出現 refusal 屬性,你可以在使用者介面顯示拒絕訊息,或在處理回應的程式碼中加入條件判斷,以處理請求遭拒的情況。
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
end模型拒絕請求時,API 回應會類似下列範例:
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,
}
},
}使用技巧與最佳做法
處理使用者提供的輸入
如果應用程式會處理使用者提供的輸入,請在提示詞中說明,當模型無法根據輸入產生有效回應時,應如何處理。
模型一律會嘗試遵循指定的結構描述,因此如果輸入與結構描述完全無關,模型可能會編造內容。
你可以在提示詞中明確要求模型,如果發現輸入與任務不相符,就傳回空的參數或指定的句子。
處理錯誤
結構化輸出的內容仍可能有誤。如果發現錯誤,可以試著調整指示、在系統指示中提供範例,或將任務拆分成較簡單的子任務。如需更多調整輸入的建議,請參閱提示工程指南。
避免 JSON 結構描述與型別不一致
為了讓 JSON 結構描述與程式語言中對應的型別保持一致,只要 SDK 有提供原生的結構描述輔助工具,我們就強烈建議使用。
如果你偏好直接指定 JSON 結構描述,可以新增 CI 規則,在有人修改 JSON 結構描述或底層資料物件時發出提醒。也可以新增 CI 步驟,根據型別定義自動產生 JSON 結構描述,或反過來從結構描述產生型別定義。
串流
當模型產生回應或函式呼叫引數時,你可以透過串流逐步處理這些內容,並將其解析為結構化資料。
這樣就能在收到完整回應前開始處理。 如果你想逐一顯示 JSON 欄位,或在收到函式呼叫引數時立即處理,串流就特別有用。
使用結構化輸出時,建議透過 SDK 處理串流。
如果想在不使用 SDK 的 stream 輔助函式的情況下,以串流方式接收函式呼叫引數,可以參考函式呼叫指南中的範例。
以下範例示範如何使用 stream 輔助函式,以串流方式接收模型回應:
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)你也可以使用 stream 輔助函式解析函式呼叫引數:
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
35
36
37
38
39
40
41
42
43from 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 == "error":
raise RuntimeError(event.message)
elif event.type in ("response.failed", "response.incomplete"):
response = event.response
message = f"Response {response.status}"
if response.error and response.error.message:
message += f": {response.error.message}"
elif response.incomplete_details and response.incomplete_details.reason:
message += f": {response.incomplete_details.reason}"
raise RuntimeError(message)
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
56
57
58
59
60
61
62
63
64
65
66
67
68require "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
raise event.message
when OpenAI::Models::Responses::ResponseFailedEvent,
OpenAI::Models::Responses::ResponseIncompleteEvent
response = event.response
detail = response.error&.message || response.incomplete_details&.reason
message = "Response #{response.status}"
message += ": #{detail}" unless detail.to_s.empty?
raise message
end
end
# Text deltas and response.output_text.done do not guarantee completion.
response = stream.get_final_response
refused = response.output.grep(OpenAI::Models::Responses::ResponseOutputMessage).any? do |message|
message.content.any? { |part| part.is_a?(OpenAI::Models::Responses::ResponseOutputRefusal) }
end
puts(refused ? "\nRefused" : "\nCompleted")支援的結構描述
結構化輸出支援 JSON Schema 語言的部分功能。
支援的型別
結構化輸出支援下列型別:
- 字串
- 數值
- 布林值
- 整數
- 物件
- 陣列
- 列舉
- anyOf
支援的屬性
除了指定屬性的型別,您也可以設定下列限制條件:
string 支援的屬性:
pattern— 字串必須符合的正規表示式。format— 字串的預先定義格式。目前支援下列格式:date-timetimedatedurationemailhostnameipv4ipv6uuid
number 支援的屬性:
multipleOf— 數值必須是此值的倍數。maximum— 數值必須小於或等於此值。exclusiveMaximum— 數值必須小於此值。minimum— 數值必須大於或等於此值。exclusiveMinimum— 數值必須大於此值。
array 支援的屬性:
minItems— 陣列的元素數量必須大於或等於此值。maxItems— 陣列的元素數量必須小於或等於此值。
以下範例示範如何使用這些型別限制:
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"
]
}
}根層級必須是物件,且不得使用 anyOf
結構描述的根層級必須是物件,且不得使用 anyOf。以 Zod 為例,在根層級使用可辨識聯集會在最上層產生 anyOf,因此無法使用下列程式碼:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19import { z } from "zod";
import { zodResponseFormat } from "openai/helpers/zod";
const BaseResponseSchema = z.object({
status: z.literal("success"),
data: z.string(),
});
const UnsuccessfulResponseSchema = z.object({
status: z.literal("error"),
error: z.string(),
});
const finalSchema = z.discriminatedUnion("status", [
BaseResponseSchema,
UnsuccessfulResponseSchema,
]);
// Invalid JSON Schema for Structured Outputs
const json = zodResponseFormat(finalSchema, "final_schema");所有欄位都必須標示為 required
若要使用結構化輸出,必須將所有欄位或函式參數標示為 required。
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"]
}
}所有欄位都必填,模型也會為每個參數傳回一個值。若要模擬選用參數,您可以使用包含 null 的聯集型別。
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"
]
}
}物件的巢狀深度與大小限制
一個結構描述中的物件屬性總數最多為 5,000 個,巢狀深度最多為 10 層。
字串總長度限制
在一個結構描述中,所有屬性名稱、定義名稱、列舉值和 const 值的字串長度合計不得超過 120,000 個字元。
列舉大小限制
一個結構描述中的所有列舉屬性合計最多可有 1,000 個列舉值。
若單一列舉屬性的值為字串,且列舉值超過 250 個,則該屬性所有列舉值的字串總長度不得超過 15,000 個字元。
每個物件都必須設定 additionalProperties: false
additionalProperties 決定物件能否包含 JSON 結構描述中未定義的額外鍵值對。
結構化輸出只支援產生已指定的鍵值對,因此開發人員必須設定 additionalProperties: false 才能啟用結構化輸出。
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"
]
}
}鍵的順序
使用結構化輸出時,輸出中的鍵會依照結構描述中各鍵的順序排列。
尚未支援的型別專用關鍵字
- 組合:
allOf、not、dependentRequired、dependentSchemas、if、then、else
微調模型也不支援下列關鍵字:
- 字串:
minLength、maxLength、pattern、format - 數值:
minimum、maximum、multipleOf - 物件:
patternProperties - 陣列:
minItems、maxItems
如果您設定 strict: true 以啟用結構化輸出,並在呼叫 API 時提供不受支援的 JSON 結構描述,API 就會傳回錯誤。
anyOf 中的每個巢狀結構描述都必須是符合此子集規範的有效 JSON 結構描述
以下是受支援的 anyOf 結構描述範例:
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"
]
}支援定義
您可以透過定義建立子結構描述,並在結構描述的各處參照這些子結構描述。以下是一個簡單的範例:
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"
}
},
"$defs": {
"step": {
"type": "object",
"properties": {
"explanation": {
"type": "string"
},
"output": {
"type": "string"
}
},
"required": [
"explanation",
"output"
],
"additionalProperties": false
}
},
"required": [
"steps",
"final_answer"
],
"additionalProperties": false
}支援遞迴結構描述
以下遞迴結構描述使用 # 參照根層級,以實現遞迴:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47{
"name": "ui",
"description": "Dynamically generated UI",
"strict": true,
"schema": {
"type": "object",
"properties": {
"type": {
"type": "string",
"description": "The type of the UI component",
"enum": ["div", "button", "header", "section", "field", "form"]
},
"label": {
"type": "string",
"description": "The label of the UI component, used for buttons or form fields"
},
"children": {
"type": "array",
"description": "Nested UI components",
"items": {
"$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
}
}以下是使用明確遞迴參照的結構描述範例:
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 模式
JSON 模式是結構化輸出功能的基本版本。JSON 模式能確保模型輸出有效的 JSON,而結構化輸出還能確保模型輸出符合你指定的結構描述。如果你的使用情境支援結構化輸出,建議使用這項功能。
啟用 JSON 模式後,模型輸出可保證是有效的 JSON,但仍有少數例外情況,應用程式應偵測並妥善處理。
若要在 Chat Completions API 中啟用 JSON 模式,請將 response_format 設為 { "type": "json_object" }。使用函式呼叫時,JSON 模式一律會啟用。
若要在 Responses API 中啟用 JSON 模式,可以將 text.format 設為 { "type": "json_object" }。使用函式呼叫時,JSON 模式一律會啟用。
重要注意事項:
- 使用 JSON 模式時,務必在對話中的至少一則訊息(例如系統訊息)中,明確要求模型產生 JSON。如果沒有這項明確指示,模型可能會持續輸出空白字元,導致請求持續執行,直到達到 Token 上限。為避免遺漏這項指示,如果上下文中沒有出現「JSON」字串,API 就會擲出錯誤。
- JSON 模式只能確保輸出是有效的 JSON,可以順利解析,無法保證輸出符合特定的結構描述。若要確保輸出符合指定的結構描述,請使用結構化輸出。如果無法使用這項功能,請使用驗證程式庫,並視需要重試,確保輸出符合預期的結構描述。
- 應用程式必須偵測並處理可能導致模型輸出不完整 JSON 物件的例外情況(詳見下文)。
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
55const 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 &&
response.choices[0].message.content !== null
) {
// 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
end資源
若想進一步瞭解結構化輸出,建議參閱下列資源:
- 參閱結構化輸出的入門實作範例。
- 瞭解如何使用結構化輸出建構多智慧體系統。