檢索 | OpenAI API

Retrieval API 讓你對資料進行語意搜尋。這項技術能找出語意相近的結果，即使只有少數關鍵字相符，甚至完全沒有相符的關鍵字。檢索本身就很實用，搭配我們的模型來整合資訊並產生回應時，更能發揮強大效用。

檢索示意圖

Retrieval API 以向量儲存庫為基礎，利用向量儲存庫為你的資料建立索引。本指南將介紹如何執行語意搜尋，並詳細說明向量儲存庫。

快速入門

建立向量儲存庫 並上傳檔案。

建立含有檔案的向量儲存庫

Python

1
2
3
4
5
6
7
8
9
10
11
12
13import OpenAI from "openai";
const client = new OpenAI();

const vector_store = await client.vectorStores.create({
  // Create vector store
  name: "Support FAQ",
});

await client.vectorStores.files.uploadAndPoll(
  vector_store.id,
  // Upload file
  fs.createReadStream("customer_policies.txt")
);

1
2
3
4
5
6
7
8
9
10
11
12from openai import OpenAI

client = OpenAI()

vector_store = client.vector_stores.create(        # Create vector store
    name="Support FAQ",
)

client.vector_stores.files.upload_and_poll(        # Upload file
    vector_store_id=vector_store.id,
    file=open("customer_policies.txt", "rb")
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30package main

import (
	"context"
	"fmt"
	"os"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	vectorStore, err := client.VectorStores.New(context.Background(), openai.VectorStoreNewParams{Name: openai.String("Support FAQ")})
	if err != nil {
		panic(err)
	}
	file, err := os.Open("customer_policies.txt")
	if err != nil {
		panic(err)
	}
	defer file.Close()
	_, err = client.VectorStores.Files.UploadAndPoll(context.Background(), vectorStore.ID, openai.FileNewParams{
		File:    openai.File(file, "customer_policies.txt", "text/plain"),
		Purpose: openai.FilePurposeAssistants,
	}, 1000)
	if err != nil {
		panic(err)
	}
	fmt.Println(vectorStore.ID)
}

1
2
3
4
5
6
7
8
9
10
11
12
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 com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.models.files.FileCreateParams;
import com.openai.models.files.FilePurpose;
import com.openai.models.vectorstores.VectorStoreCreateParams;
import com.openai.models.vectorstores.files.FileRetrieveParams;
import com.openai.models.vectorstores.files.VectorStoreFile;
import java.nio.file.Path;

var store =
    client.vectorStores().create(VectorStoreCreateParams.builder().name("Support FAQ").build());
var uploaded =
    client
        .files()
        .create(
            FileCreateParams.builder()
                .file(Path.of(System.getenv("OPENAI_EXAMPLE_FILE_PATH")))
                .purpose(FilePurpose.ASSISTANTS)
                .build());
var file =
    client
        .vectorStores()
        .files()
        .create(
            store.id(),
            com.openai.models.vectorstores.files.FileCreateParams.builder()
                .fileId(uploaded.id())
                .build());
while (file.status().equals(VectorStoreFile.Status.IN_PROGRESS)) {
  Thread.sleep(1000);
  file =
      client
          .vectorStores()
          .files()
          .retrieve(file.id(), FileRetrieveParams.builder().vectorStoreId(store.id()).build());
}
System.out.println(store.id());

1
2
3
4
5
6
7
8
9
10
11
12
13require "openai"
require "pathname"

client = OpenAI::Client.new
store = client.vector_stores.create(name: "Support FAQ")
file = client.vector_stores.files.upload_and_poll(
  store.id,
  file: Pathname("customer_policies.txt"),
  timeout: 600
)
raise "File ingestion ended with status: #{file.status}" unless file.status == OpenAI::VectorStores::VectorStoreFile::Status::COMPLETED

puts(store.id)

傳送搜尋查詢 以取得相關結果。

搜尋查詢

Python

1
2
3
4
5const userQuery = "What is the return policy?";

const results = await client.vectorStores.search(vector_store.id, {
  query: userQuery,
});

1
2
3
4
5
6user_query = "What is the return policy?"

results = client.vector_stores.search(
    vector_store_id=vector_store.id,
    query=user_query,
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19package main

import (
	"context"
	"fmt"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	results, err := client.VectorStores.Search(context.Background(), "vs_123", openai.VectorStoreSearchParams{
		Query: openai.VectorStoreSearchParamsQueryUnion{OfString: openai.String("What is the return policy?")},
	})
	if err != nil {
		panic(err)
	}
	fmt.Println(results.Data)
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.models.vectorstores.VectorStoreSearchParams;

String vectorStoreId = "vs_123";

var results =
    client
        .vectorStores()
        .search(
            vectorStoreId,
            VectorStoreSearchParams.builder().query("What is the return policy?").build());

System.out.println(results.data());

1
2
3
4
5require "openai"

client = OpenAI::Client.new
results = client.vector_stores.search("vs_123", query: "What is the return policy?")
puts(results.data&.first&.content)

如要瞭解如何將搜尋結果搭配我們的模型使用，請參閱整合資訊並產生回應一節。

語意搜尋

語意搜尋 利用嵌入向量找出語意相關的結果。這項技術的關鍵在於，即使結果只有少數相同的關鍵字，甚至完全沒有相同的關鍵字，也能被找出來，而傳統搜尋技術可能會遺漏這些結果。

例如，來看看 "When did we go to the moon?" 可能會得到哪些結果：

文字	關鍵字相似度	語意相似度
首次登月發生於 1969 年 7 月。	0%	65%
第一位登上月球的人是尼爾．阿姆斯壯。	27%	43%
我吃了那個月餅，覺得很好吃。	40%	28%

（關鍵字相似度使用交集與聯集的比值計算；語意相似度則搭配 text-embedding-3-small，使用餘弦相似度計算。）

請注意，最相關的結果完全不含搜尋查詢中的任何字詞。這種彈性讓語意搜尋成為查詢各種規模知識庫的強大技術。

語意搜尋以向量儲存庫為基礎，本指南稍後會詳細介紹向量儲存庫。本節將著重說明語意搜尋的運作方式。

執行語意搜尋

你可以使用 search 函式，並以自然語言指定 query，來查詢向量儲存庫。這會傳回結果清單，每筆結果都包含相關的文字區塊、相似度分數及來源檔案。

搜尋查詢

Python

1
2
3const results = await client.vectorStores.search(vector_store.id, {
  query: "How many woodchucks are allowed per passenger?",
});

1
2
3
4results = client.vector_stores.search(
    vector_store_id=vector_store.id,
    query="How many woodchucks are allowed per passenger?",
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19package main

import (
	"context"
	"fmt"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	results, err := client.VectorStores.Search(context.Background(), "vs_123", openai.VectorStoreSearchParams{
		Query: openai.VectorStoreSearchParamsQueryUnion{OfString: openai.String("How many woodchucks are allowed per passenger?")},
	})
	if err != nil {
		panic(err)
	}
	fmt.Println(results.Data)
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.models.vectorstores.VectorStoreSearchParams;

String vectorStoreId = "vs_123";

var results =
    client
        .vectorStores()
        .search(
            vectorStoreId,
            VectorStoreSearchParams.builder()
                .query("How many woodchucks are allowed per passenger?")
                .build());

System.out.println(results.data());

1
2
3
4
5
6
7
8require "openai"

client = OpenAI::Client.new
results = client.vector_stores.search(
  "vs_123",
  query: "How many woodchucks are allowed per passenger?"
)
puts(results.data&.first&.content)

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
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42{
  "object": "vector_store.search_results.page",
  "search_query": "How many woodchucks are allowed per passenger?",
  "data": [
    {
      "file_id": "file-12345",
      "filename": "woodchuck_policy.txt",
      "score": 0.85,
      "attributes": {
        "region": "North America",
        "author": "Wildlife Department"
      },
      "content": [
        {
          "type": "text",
          "text": "According to the latest regulations, each passenger is allowed to carry up to two woodchucks."
        },
        {
          "type": "text",
          "text": "Ensure that the woodchucks are properly contained during transport."
        }
      ]
    },
    {
      "file_id": "file-67890",
      "filename": "transport_guidelines.txt",
      "score": 0.75,
      "attributes": {
        "region": "North America",
        "author": "Transport Authority"
      },
      "content": [
        {
          "type": "text",
          "text": "Passengers must adhere to the guidelines set forth by the Transport Authority regarding the transport of woodchucks."
        }
      ]
    }
  ],
  "has_more": false,
  "next_page": null
}

回應預設最多包含 10 筆結果，但你可以透過 max_num_results 參數將上限設為最多 50 筆。

查詢改寫

某些查詢方式能得到更好的結果，因此我們提供了一項設定，可自動改寫查詢以達到最佳效果。執行 search 時，設定 rewrite_query=true 即可啟用這項功能。

改寫後的查詢會出現在結果的 search_query 欄位中。

原始查詢	改寫後的查詢
我想知道主要辦公大樓的高度。	主要辦公大樓高度
運送危險物品有哪些安全規範？	危險物品安全規範
我要如何針對服務問題提出申訴？	服務申訴提出流程

屬性篩選

屬性篩選可透過套用條件來縮小結果範圍，例如將搜尋限制在特定日期範圍內。你可以在 attribute_filter 中定義及組合條件，在執行語意搜尋前，先依檔案屬性篩選出目標檔案。

使用 比較篩選器 ，將檔案 attributes 中的特定 key 與指定的 value 比較；使用 複合篩選器 ，則可透過 and 和 or 組合多個篩選器。

比較篩選器

1
2
3
4
5{
  "type": "eq" | "ne" | "gt" | "gte" | "lt" | "lte" | "in" | "nin",  // comparison operators
  "key": "attributes_key",                           // attributes key
  "value": "target_value"                             // value to compare against
}

複合篩選器

1
2
3
4{
  "type": "and" | "or",                                // logical operators
  "filters": [...]
}

以下是一些篩選器範例。

依地區篩選

1
2
3
4
5{
  "type": "eq",
  "key": "region",
  "value": "us"
}

依日期範圍篩選

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15{
  "type": "and",
  "filters": [
    {
      "type": "gte",
      "key": "date",
      "value": 1704067200  // unix timestamp for 2024-01-01
    },
    {
      "type": "lte",
      "key": "date",
      "value": 1710892800  // unix timestamp for 2024-03-20
    }
  ]
}

排序

如果檔案搜尋結果的相關性不夠高，你可以調整 ranking_options，改善回應品質。可調整的項目包括指定 ranker，例如 auto 或 default-2024-08-21，以及將 score_threshold 設為 0.0 到 1.0 之間的值。提高 score_threshold 會讓結果僅包含相關性較高的區塊，但也可能排除一些有用的區塊。提供 ranking_options.hybrid_search 時，你也可以調整 hybrid_search.embedding_weight（rrf_embedding_weight）和 hybrid_search.text_weight（rrf_text_weight），控制倒數排名融合如何平衡語意嵌入比對與稀疏關鍵字比對的結果。提高前者的權重可強調語意相似度，提高後者的權重可強調文字重疊程度，且至少須有一個權重大於零。

向量儲存庫

向量儲存庫是為 Retrieval API 和檔案搜尋工具提供語意搜尋功能的容器。將檔案加入向量儲存庫後，系統會自動將檔案分割成區塊、產生嵌入向量，並建立索引。

向量儲存庫包含 vector_store_file 物件，每個物件都以一個 file 物件為基礎。

物件類型	說明
`file`	代表透過 Files API 上傳的內容。通常與向量儲存庫搭配使用，也可用於微調和其他使用案例。
`vector_store`	存放可搜尋檔案的容器。
`vector_store.file`	封裝類型，專門用來代表已分割成區塊、產生嵌入向量，並與 `vector_store` 建立關聯的 `file`。包含用於篩選的 `attributes` 對應表。

定價

費用依所有向量儲存庫使用的總儲存空間計算，儲存用量取決於解析後的區塊及其對應嵌入向量的大小。

儲存空間	費用
1 GB 以內（所有儲存庫合計）	免費
超過 1 GB 的部分	$0.10/GB/天

如需瞭解盡量降低費用的選項，請參閱到期政策。

向量儲存庫操作

建立向量儲存庫

Python

1
2
3
4await client.vectorStores.create({
  name: "Support FAQ",
  file_ids: ["file_123"],
});

1
2
3
4client.vector_stores.create(
    name="Support FAQ",
    file_ids=["file_123"]
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20package main

import (
	"context"
	"fmt"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	vectorStore, err := client.VectorStores.New(context.Background(), openai.VectorStoreNewParams{
		Name:    openai.String("Support FAQ"),
		FileIDs: []string{"file_123"},
	})
	if err != nil {
		panic(err)
	}
	fmt.Println(vectorStore.ID)
}

1
2
3
4
5
6
7
8
9
10
11
12
13import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.models.vectorstores.VectorStoreCreateParams;

String fileId = "file_123";

var store =
    client
        .vectorStores()
        .create(
            VectorStoreCreateParams.builder().name("Support FAQ").addFileId(fileId).build());

System.out.println(store.id());

1
2
3
4
5
6
7
8require "openai"

client = OpenAI::Client.new
store = client.vector_stores.create(
  name: "Support FAQ",
  file_ids: ["file_123"]
)
puts(store.id)

取得向量儲存庫

Python

1await client.vectorStores.retrieve("vs_123");

1
2
3client.vector_stores.retrieve(
    vector_store_id="vs_123"
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17package main

import (
	"context"
	"fmt"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	vectorStore, err := client.VectorStores.Get(context.Background(), "vs_123")
	if err != nil {
		panic(err)
	}
	fmt.Println(vectorStore.ID)
}

1
2
3
4
5
6import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;

String vectorStoreId = "vs_123";

System.out.println(client.vectorStores().retrieve(vectorStoreId).id());

1
2
3
4
5require "openai"

client = OpenAI::Client.new
store = client.vector_stores.retrieve("vs_123")
puts(store.id)

更新向量儲存庫

Python

1
2
3await client.vectorStores.update("vs_123", {
  name: "Support FAQ Updated",
});

1
2
3
4client.vector_stores.update(
    vector_store_id="vs_123",
    name="Support FAQ Updated"
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19package main

import (
	"context"
	"fmt"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	vectorStore, err := client.VectorStores.Update(context.Background(), "vs_123", openai.VectorStoreUpdateParams{
		Name: openai.String("Support FAQ Updated"),
	})
	if err != nil {
		panic(err)
	}
	fmt.Println(vectorStore.Name)
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.models.vectorstores.VectorStoreUpdateParams;

String vectorStoreId = "vs_123";

var store =
    client
        .vectorStores()
        .update(
            vectorStoreId,
            VectorStoreUpdateParams.builder().name("Updated knowledge base").build());

System.out.println(store.name());

1
2
3
4
5require "openai"

client = OpenAI::Client.new
store = client.vector_stores.update("vs_123", name: "Updated knowledge base")
puts(store.name)

刪除向量儲存庫

Python

1await client.vectorStores.delete("vs_123");

1
2
3client.vector_stores.delete(
    vector_store_id="vs_123"
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17package main

import (
	"context"
	"fmt"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	deleted, err := client.VectorStores.Delete(context.Background(), "vs_123")
	if err != nil {
		panic(err)
	}
	fmt.Println(deleted.Deleted)
}

1
2
3
4
5
6import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;

String vectorStoreId = "vs_123";

System.out.println(client.vectorStores().delete(vectorStoreId).deleted());

1
2
3
4
5require "openai"

client = OpenAI::Client.new
deleted = client.vector_stores.delete("vs_123")
puts(deleted.deleted)

列出向量儲存庫

Python

await client.vectorStores.list();

client.vector_stores.list()

package main

import (
	"context"
	"fmt"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	vectorStores, err := client.VectorStores.List(context.Background(), openai.VectorStoreListParams{})
	if err != nil {
		panic(err)
	}
	fmt.Println(vectorStores.Data)
}

import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;

System.out.println(client.vectorStores().list().data());

require "openai"

client = OpenAI::Client.new
stores = client.vector_stores.list(limit: 10)
puts((stores.data || []).length)

向量儲存庫檔案操作

某些操作（例如 vector_store.file 的 create）採非同步方式執行，可能需要一些時間才能完成。您可以使用我們的輔助函式（例如 create_and_poll）阻塞等待，直到操作完成，也可以自行檢查狀態。從向量儲存庫移除檔案採用最終一致性，因此搜尋結果在短時間內仍可能包含已移除檔案的內容。

新增檔案的速率限制以各個向量儲存庫 ID 為單位。對 /vector_stores/{vector_store_id}/files 和 /vector_stores/{vector_store_id}/file_batches 發出的請求，共用每個向量儲存庫每分鐘 300 次請求的限制。

建立向量儲存庫檔案

Python

1
2
3await client.vectorStores.files.createAndPoll("vs_123", {
  file_id: "file_123",
});

1
2
3
4client.vector_stores.files.create_and_poll(
    vector_store_id="vs_123",
    file_id="file_123"
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19package main

import (
	"context"
	"fmt"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	file, err := client.VectorStores.Files.NewAndPoll(context.Background(), "vs_123", openai.VectorStoreFileNewParams{
		FileID: "file_123",
	}, 1000)
	if err != nil {
		panic(err)
	}
	fmt.Println(file.ID)
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.models.vectorstores.files.FileCreateParams;

String vectorStoreId = "vs_123";

String fileId = "file_123";

var file =
    client
        .vectorStores()
        .files()
        .create(vectorStoreId, FileCreateParams.builder().fileId(fileId).build());

System.out.println(file.id());

1
2
3
4
5require "openai"

client = OpenAI::Client.new
file = client.vector_stores.files.create("vs_123", file_id: "file_123")
puts(file.id)

上傳向量儲存庫檔案

Python

1
2
3
4await client.vectorStores.files.uploadAndPoll(
  "vs_123",
  fs.createReadStream("customer_policies.txt")
);

1
2
3
4client.vector_stores.files.upload_and_poll(
    vector_store_id="vs_123",
    file=open("customer_policies.txt", "rb")
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26package main

import (
	"context"
	"fmt"
	"os"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	file, err := os.Open("customer_policies.txt")
	if err != nil {
		panic(err)
	}
	defer file.Close()
	result, err := client.VectorStores.Files.UploadAndPoll(context.Background(), "vs_123", openai.FileNewParams{
		File:    openai.File(file, "customer_policies.txt", "text/plain"),
		Purpose: openai.FilePurposeAssistants,
	}, 1000)
	if err != nil {
		panic(err)
	}
	fmt.Println(result.ID)
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.models.files.FileCreateParams;
import com.openai.models.files.FilePurpose;
import com.openai.models.vectorstores.files.FileRetrieveParams;
import com.openai.models.vectorstores.files.VectorStoreFile;
import java.nio.file.Path;

String vectorStoreId = "vs_123";
var uploaded =
    client
        .files()
        .create(
            FileCreateParams.builder()
                .file(Path.of(System.getenv("OPENAI_EXAMPLE_FILE_PATH")))
                .purpose(FilePurpose.ASSISTANTS)
                .build());
var file =
    client
        .vectorStores()
        .files()
        .create(
            vectorStoreId,
            com.openai.models.vectorstores.files.FileCreateParams.builder()
                .fileId(uploaded.id())
                .build());
while (file.status().equals(VectorStoreFile.Status.IN_PROGRESS)) {
  Thread.sleep(1000);
  file =
      client
          .vectorStores()
          .files()
          .retrieve(
              file.id(), FileRetrieveParams.builder().vectorStoreId(vectorStoreId).build());
}
System.out.println(file.id());

1
2
3
4
5
6
7
8
9
10
11
12require "openai"
require "pathname"

client = OpenAI::Client.new
vector_store_file = client.vector_stores.files.upload_and_poll(
  "vs_123",
  file: Pathname("customer_policies.txt"),
  timeout: 600
)
raise "File ingestion ended with status: #{vector_store_file.status}" unless vector_store_file.status == OpenAI::VectorStores::VectorStoreFile::Status::COMPLETED

puts(vector_store_file.id)

取得向量儲存庫檔案

Python

1
2
3await client.vectorStores.files.retrieve("file_123", {
  vector_store_id: "vs_123",
});

1
2
3
4client.vector_stores.files.retrieve(
    vector_store_id="vs_123",
    file_id="file_123"
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17package main

import (
	"context"
	"fmt"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	file, err := client.VectorStores.Files.Get(context.Background(), "vs_123", "file_123")
	if err != nil {
		panic(err)
	}
	fmt.Println(file.ID)
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;

String fileId = "file_123";

String vectorStoreId = "vs_123";

System.out.println(
    client
        .vectorStores()
        .files()
        .retrieve(
            fileId,
            com.openai.models.vectorstores.files.FileRetrieveParams.builder()
                .vectorStoreId(vectorStoreId)
                .build())
        .id());

1
2
3
4
5require "openai"

client = OpenAI::Client.new
file = client.vector_stores.files.retrieve("file_123", vector_store_id: "vs_123")
puts(file.id)

更新向量儲存庫檔案

Python

1
2
3
4await client.vectorStores.files.update("file_123", {
  vector_store_id: "vs_123",
  attributes: { key: "value" },
});

1
2
3
4
5client.vector_stores.files.update(
    vector_store_id="vs_123",
    file_id="file_123",
    attributes={"key": "value"}
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21package main

import (
	"context"
	"fmt"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	file, err := client.VectorStores.Files.Update(context.Background(), "vs_123", "file_123", openai.VectorStoreFileUpdateParams{
		Attributes: map[string]openai.VectorStoreFileUpdateParamsAttributeUnion{
			"key": {OfString: openai.String("value")},
		},
	})
	if err != nil {
		panic(err)
	}
	fmt.Println(file.ID)
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.vectorstores.files.FileUpdateParams;

String fileId = "file_123";

String vectorStoreId = "vs_123";

var file =
    client
        .vectorStores()
        .files()
        .update(
            fileId,
            FileUpdateParams.builder()
                .vectorStoreId(vectorStoreId)
                .attributes(
                    FileUpdateParams.Attributes.builder()
                        .putAdditionalProperty("category", JsonValue.from("policy"))
                        .build())
                .build());

System.out.println(file.id());

1
2
3
4
5require "openai"

client = OpenAI::Client.new
file = client.vector_stores.files.update("file_123", vector_store_id: "vs_123", attributes: { category: "policy" })
puts(file.id)

刪除向量儲存庫檔案

Python

1
2
3await client.vectorStores.files.delete("file_123", {
  vector_store_id: "vs_123",
});

1
2
3
4client.vector_stores.files.delete(
    vector_store_id="vs_123",
    file_id="file_123"
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17package main

import (
	"context"
	"fmt"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	deleted, err := client.VectorStores.Files.Delete(context.Background(), "vs_123", "file_123")
	if err != nil {
		panic(err)
	}
	fmt.Println(deleted.Deleted)
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;

String fileId = "file_123";

String vectorStoreId = "vs_123";

System.out.println(
    client
        .vectorStores()
        .files()
        .delete(
            fileId,
            com.openai.models.vectorstores.files.FileDeleteParams.builder()
                .vectorStoreId(vectorStoreId)
                .build())
        .deleted());

1
2
3
4
5require "openai"

client = OpenAI::Client.new
deleted = client.vector_stores.files.delete("file_123", vector_store_id: "vs_123")
puts(deleted.deleted)

列出向量儲存庫檔案

Python

1await client.vectorStores.files.list("vs_123");

1
2
3client.vector_stores.files.list(
    vector_store_id="vs_123"
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17package main

import (
	"context"
	"fmt"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	files, err := client.VectorStores.Files.List(context.Background(), "vs_123", openai.VectorStoreFileListParams{})
	if err != nil {
		panic(err)
	}
	fmt.Println(files.Data)
}

1
2
3
4
5
6import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;

String vectorStoreId = "vs_123";

System.out.println(client.vectorStores().files().list(vectorStoreId).data());

1
2
3
4
5require "openai"

client = OpenAI::Client.new
files = client.vector_stores.files.list("vs_123")
puts((files.data || []).length)

批次處理操作

建立批次處理的操作

Python

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18await client.vectorStores.fileBatches.createAndPoll("vs_123", {
  files: [
    {
      file_id: "file_123",
      attributes: { department: "finance" },
    },
    {
      file_id: "file_456",
      chunking_strategy: {
        type: "static",
        static: {
          max_chunk_size_tokens: 1200,
          chunk_overlap_tokens: 200,
        },
      },
    },
  ],
});

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17client.vector_stores.file_batches.create_and_poll(
    vector_store_id="vs_123",
    files=[
        {
            "file_id": "file_123",
            "attributes": {"department": "finance"}
        },
        {
            "file_id": "file_456",
            "chunking_strategy": {
                "type": "static",
                "max_chunk_size_tokens": 1200,
                "chunk_overlap_tokens": 200
            }
        }
    ]
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32package main

import (
	"context"
	"fmt"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	batch, err := client.VectorStores.FileBatches.NewAndPoll(context.Background(), "vs_123", openai.VectorStoreFileBatchNewParams{
		Files: []openai.VectorStoreFileBatchNewParamsFile{
			{
				FileID: "file_123",
				Attributes: map[string]openai.VectorStoreFileBatchNewParamsFileAttributeUnion{
					"department": {OfString: openai.String("finance")},
				},
			},
			{
				FileID: "file_456",
				ChunkingStrategy: openai.FileChunkingStrategyParamUnion{OfStatic: &openai.StaticFileChunkingStrategyObjectParam{
					Static: openai.StaticFileChunkingStrategyParam{MaxChunkSizeTokens: 1200, ChunkOverlapTokens: 200},
				}},
			},
		},
	}, 1000)
	if err != nil {
		panic(err)
	}
	fmt.Println(batch.ID)
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14
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
47import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.vectorstores.StaticFileChunkingStrategy;
import com.openai.models.vectorstores.filebatches.FileBatchCreateParams;
import com.openai.models.vectorstores.filebatches.FileBatchRetrieveParams;
import com.openai.models.vectorstores.filebatches.VectorStoreFileBatch;

String vectorStoreId = "vs_123";
String fileId = "file_123";
String fileId2 = "file_456";
var first =
    FileBatchCreateParams.File.builder()
        .fileId(fileId)
        .attributes(
            FileBatchCreateParams.File.Attributes.builder()
                .putAdditionalProperty("department", JsonValue.from("finance"))
                .build())
        .build();
var second =
    FileBatchCreateParams.File.builder()
        .fileId(fileId2)
        .staticChunkingStrategy(
            StaticFileChunkingStrategy.builder()
                .maxChunkSizeTokens(1200)
                .chunkOverlapTokens(200)
                .build())
        .build();

var batch =
    client
        .vectorStores()
        .fileBatches()
        .create(
            vectorStoreId,
            FileBatchCreateParams.builder().addFile(first).addFile(second).build());
while (batch.status().equals(VectorStoreFileBatch.Status.IN_PROGRESS)) {
  Thread.sleep(1000);
  batch =
      client
          .vectorStores()
          .fileBatches()
          .retrieve(
              batch.id(),
              FileBatchRetrieveParams.builder().vectorStoreId(vectorStoreId).build());
}
System.out.println(batch.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
33
34
35
36
37require "openai"

client = OpenAI::Client.new
batch = client.vector_stores.file_batches.create_and_poll(
  "vs_123",
  files: [
    {
      file_id: "file_123",
      attributes: { department: "finance" }
    },
    {
      file_id: "file_456",
      chunking_strategy: {
        type: :static,
        static: {
          max_chunk_size_tokens: 1_200,
          chunk_overlap_tokens: 200
        }
      }
    }
  ],
  timeout: 600
)
raise "File ingestion ended with status: #{batch.status}" unless batch.status == OpenAI::VectorStores::VectorStoreFileBatch::Status::COMPLETED

raise "File ingestion failed for #{batch.file_counts.failed} file(s)" if batch.file_counts.failed.positive?

# Live validation of per-file batches returned default chunking despite overrides.
file = client.vector_stores.files.retrieve("file_456", vector_store_id: "vs_123")
strategy = file.chunking_strategy
unless strategy.is_a?(OpenAI::StaticFileChunkingStrategyObject) &&
       strategy.static.max_chunk_size_tokens == 1_200 &&
       strategy.static.chunk_overlap_tokens == 200
  raise "Requested chunking was not applied to #{file.id}: #{strategy.to_json}"
end

puts(batch.status)

取得批次處理的操作

Python

1
2
3await client.vectorStores.fileBatches.retrieve("vsfb_123", {
  vector_store_id: "vs_123",
});

1
2
3
4client.vector_stores.file_batches.retrieve(
    vector_store_id="vs_123",
    batch_id="vsfb_123"
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17package main

import (
	"context"
	"fmt"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	batch, err := client.VectorStores.FileBatches.Get(context.Background(), "vs_123", "vsfb_123")
	if err != nil {
		panic(err)
	}
	fmt.Println(batch.ID)
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;

String fileBatchId = "vsfb_123";

String vectorStoreId = "vs_123";

System.out.println(
    client
        .vectorStores()
        .fileBatches()
        .retrieve(
            fileBatchId,
            com.openai.models.vectorstores.filebatches.FileBatchRetrieveParams.builder()
                .vectorStoreId(vectorStoreId)
                .build())
        .status());

1
2
3
4
5
6
7
8require "openai"

client = OpenAI::Client.new
batch = client.vector_stores.file_batches.retrieve(
  "vsfb_123",
  vector_store_id: "vs_123"
)
puts(batch.status)

取消批次處理的操作

Python

1
2
3await client.vectorStores.fileBatches.cancel("vsfb_123", {
  vector_store_id: "vs_123",
});

1
2
3
4client.vector_stores.file_batches.cancel(
    vector_store_id="vs_123",
    batch_id="vsfb_123"
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17package main

import (
	"context"
	"fmt"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	batch, err := client.VectorStores.FileBatches.Cancel(context.Background(), "vs_123", "vsfb_123")
	if err != nil {
		panic(err)
	}
	fmt.Println(batch.Status)
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;

String fileBatchId = "vsfb_123";

String vectorStoreId = "vs_123";

System.out.println(
    client
        .vectorStores()
        .fileBatches()
        .cancel(
            fileBatchId,
            com.openai.models.vectorstores.filebatches.FileBatchCancelParams.builder()
                .vectorStoreId(vectorStoreId)
                .build())
        .status());

1
2
3
4
5
6
7
8require "openai"

client = OpenAI::Client.new
batch = client.vector_stores.file_batches.cancel(
  "vsfb_123",
  vector_store_id: "vs_123"
)
puts(batch.status)

列出批次處理中的檔案

Python

1
2
3await client.vectorStores.fileBatches.listFiles("vsfb_123", {
  vector_store_id: "vs_123",
});

1
2
3
4client.vector_stores.file_batches.list_files(
    "vsfb_123",
    vector_store_id="vs_123"
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17package main

import (
	"context"
	"fmt"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	files, err := client.VectorStores.FileBatches.ListFiles(context.Background(), "vs_123", "vsfb_123", openai.VectorStoreFileBatchListFilesParams{})
	if err != nil {
		panic(err)
	}
	fmt.Println(files.Data)
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;

String fileBatchId = "vsfb_123";

String vectorStoreId = "vs_123";

System.out.println(
    client
        .vectorStores()
        .fileBatches()
        .listFiles(
            fileBatchId,
            com.openai.models.vectorstores.filebatches.FileBatchListFilesParams.builder()
                .vectorStoreId(vectorStoreId)
                .build())
        .data());

1
2
3
4
5
6
7
8require "openai"

client = OpenAI::Client.new
files = client.vector_stores.file_batches.list_files(
  "vsfb_123",
  vector_store_id: "vs_123"
)
puts((files.data || []).length)

建立批次時，你可以提供 file_ids，並選擇性地搭配 attributes 和／或 chunking_strategy；也可以使用 files 陣列，為每個檔案傳入包含 file_id 以及選填的 attributes 和 chunking_strategy 的物件。這兩種方式不能同時使用，讓你能明確控制所有檔案是否共用相同設定，或是針對個別檔案覆寫設定。

若要提高將資料匯入單一向量儲存庫的吞吐量，建議盡可能使用批次建立。每個批次可在一次請求中包含最多 500 個檔案。相較於傳送大量單一檔案建立請求，這通常能減少資源爭用，並降低端到端延遲。

屬性

每個 vector_store.file 都可以有關聯的 attributes，這是一個儲存值的字典，可在執行搭配屬性篩選的語意搜尋時參照。字典最多可有 16 個鍵，每個鍵的長度上限為 256 個字元。

建立具有屬性的向量儲存庫檔案

Python

1
2
3
4
5
6
7
8await client.vectorStores.files.create("<vector_store_id>", {
  file_id: "file_123",
  attributes: {
    region: "US",
    category: "Marketing",
    date: 1672531200, // Jan 1, 2023
  },
});

1
2
3
4
5
6
7
8
9client.vector_stores.files.create(
    vector_store_id="<vector_store_id>",
    file_id="file_123",
    attributes={
        "region": "US",
        "category": "Marketing",
        "date": 1672531200      # Jan 1, 2023
    }
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24package main

import (
	"context"
	"fmt"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	file, err := client.VectorStores.Files.New(context.Background(), "<vector_store_id>", openai.VectorStoreFileNewParams{
		FileID: "file_123",
		Attributes: map[string]openai.VectorStoreFileNewParamsAttributeUnion{
			"region":   {OfString: openai.String("US")},
			"category": {OfString: openai.String("Marketing")},
			"date":     {OfFloat: openai.Float(1672531200)},
		},
	})
	if err != nil {
		panic(err)
	}
	fmt.Println(file.ID)
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.vectorstores.files.FileCreateParams;

String vectorStoreId = "<vector_store_id>";

String fileId = "file_123";

var file =
    client
        .vectorStores()
        .files()
        .create(
            vectorStoreId,
            FileCreateParams.builder()
                .fileId(fileId)
                .attributes(
                    FileCreateParams.Attributes.builder()
                        .putAdditionalProperty("category", JsonValue.from("policy"))
                        .build())
                .build());

System.out.println(file.id());

1
2
3
4
5require "openai"

client = OpenAI::Client.new
file = client.vector_stores.files.create("<vector_store_id>", file_id: "file_123", attributes: { category: "policy" })
puts(file.id)

到期政策

你可以使用 expires_after 為 vector_store 物件設定到期政策。向量儲存庫到期後，所有關聯的 vector_store.file 物件都會遭到刪除，也不再產生費用。

設定向量儲存庫的到期政策

Python

1
2
3
4
5
6await client.vectorStores.update("vs_123", {
  expires_after: {
    anchor: "last_active_at",
    days: 7,
  },
});

1
2
3
4
5
6
7client.vector_stores.update(
    vector_store_id="vs_123",
    expires_after={
        "anchor": "last_active_at",
        "days": 7
    }
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19package main

import (
	"context"
	"fmt"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	vectorStore, err := client.VectorStores.Update(context.Background(), "vs_123", openai.VectorStoreUpdateParams{
		ExpiresAfter: openai.VectorStoreUpdateParamsExpiresAfter{Days: 7},
	})
	if err != nil {
		panic(err)
	}
	fmt.Println(vectorStore.ExpiresAfter)
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.core.JsonValue;
import com.openai.models.vectorstores.VectorStoreUpdateParams;

String vectorStoreId = "vs_123";

var store =
    client
        .vectorStores()
        .update(
            vectorStoreId,
            VectorStoreUpdateParams.builder()
                .expiresAfter(
                    VectorStoreUpdateParams.ExpiresAfter.builder()
                        .anchor(JsonValue.from("last_active_at"))
                        .days(7)
                        .build())
                .build());

System.out.println(store.expiresAfter().orElseThrow());

1
2
3
4
5
6
7
8
9
10
11require "openai"

client = OpenAI::Client.new
store = client.vector_stores.update(
  "vs_123",
  expires_after: {
    anchor: :last_active_at,
    days: 7
  }
)
puts(store.expires_after)

限制

檔案大小上限為 512 MB。每個檔案應包含不超過 5,000,000 個 Token（附加檔案時會自動計算）。

分塊

預設情況下，max_chunk_size_tokens 設為 800，chunk_overlap_tokens 設為 400。這表示每個檔案在建立索引時，都會分割成各含 800 個 Token 的區塊，相鄰區塊之間有 400 個 Token 重疊。

你可以在將檔案新增至向量儲存庫時，設定 chunking_strategy 來調整分塊方式。此策略有以下限制：

max_chunk_size_tokens 必須介於 100 到 4096 之間，包含上下限。
chunk_overlap_tokens 必須為非負值，且不應超過 max_chunk_size_tokens / 2。

對於 text/ MIME 類型，編碼必須是 utf-8、utf-16 或 ascii 其中之一。

檔案格式	MIME 類型
`.c`	`text/x-c`
`.cpp`	`text/x-c++`
`.cs`	`text/x-csharp`
`.css`	`text/css`
`.doc`	`application/msword`
`.docx`	`application/vnd.openxmlformats-officedocument.wordprocessingml.document`
`.go`	`text/x-golang`
`.html`	`text/html`
`.java`	`text/x-java`
`.js`	`text/javascript`
`.json`	`application/json`
`.md`	`text/markdown`
`.pdf`	`application/pdf`
`.php`	`text/x-php`
`.pptx`	`application/vnd.openxmlformats-officedocument.presentationml.presentation`
`.py`	`text/x-python`
`.py`	`text/x-script.python`
`.rb`	`text/x-ruby`
`.sh`	`application/x-sh`
`.tex`	`text/x-tex`
`.ts`	`application/typescript`
`.txt`	`text/plain`

整合產生回應

執行查詢後，你可能會想根據結果整合產生回應。你可以將結果和原始查詢提供給我們的模型，讓模型產生有依據的回應。

執行搜尋查詢以取得結果

Python

1
2
3
4
5
6
7
8
9import OpenAI from "openai";

const client = new OpenAI();

const userQuery = "What is the return policy?";

const results = await client.vectorStores.search(vector_store.id, {
  query: userQuery,
});

1
2
3
4
5
6
7
8
9
10from openai import OpenAI

client = OpenAI()

user_query = "What is the return policy?"

results = client.vector_stores.search(
    vector_store_id=vector_store.id,
    query=user_query,
)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19package main

import (
	"context"
	"fmt"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	results, err := client.VectorStores.Search(context.Background(), "vs_123", openai.VectorStoreSearchParams{
		Query: openai.VectorStoreSearchParamsQueryUnion{OfString: openai.String("What is the return policy?")},
	})
	if err != nil {
		panic(err)
	}
	fmt.Println(results.Data)
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.models.vectorstores.VectorStoreSearchParams;

String vectorStoreId = "vs_123";

var results =
    client
        .vectorStores()
        .search(
            vectorStoreId,
            VectorStoreSearchParams.builder().query("What is the return policy?").build());

System.out.println(results.data());

1
2
3
4
5
6
7
8require "openai"

client = OpenAI::Client.new
results = client.vector_stores.search(
  "vs_123",
  query: "What is the return policy?"
)
puts(results.data)

根據結果整合產生回應

Python

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22const formattedResults = formatResults(results.data);
// Join the text content of all results
const textSources = results.data
  .map((result) => result.content.map((c) => c.text).join("\n"))
  .join("\n");

const completion = await client.chat.completions.create({
  model: "gpt-6-astra",
  messages: [
    {
      role: "developer",
      content:
        "Produce a concise answer to the query based on the provided sources.",
    },
    {
      role: "user",
      content: `Sources: ${formattedResults}\n\nQuery: '${userQuery}'`,
    },
  ],
});

console.log(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# Use results and user_query from the preceding search step.
formatted_results = format_results(results.data)

"\n".join("\n".join(c.text for c in result.content) for result in results.data)

completion = client.chat.completions.create(
    model="gpt-6-astra",
    messages=[
        {
            "role": "developer",
            "content": "Produce a concise answer to the query based on the provided sources.",
        },
        {
            "role": "user",
            "content": f"Sources: {formatted_results}\n\nQuery: '{user_query}'",
        },
    ],
)

print(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
46package main

import (
	"context"
	"fmt"
	"strings"

	"github.com/openai/openai-go/v3"
)

func main() {
	client := openai.NewClient()
	userQuery := "What is the return policy?"
	results, err := client.VectorStores.Search(context.Background(), "vs_123", openai.VectorStoreSearchParams{
		Query: openai.VectorStoreSearchParamsQueryUnion{OfString: openai.String(userQuery)},
	})
	if err != nil {
		panic(err)
	}

	completion, err := client.Chat.Completions.New(context.Background(), openai.ChatCompletionNewParams{
		Model: "gpt-6-astra",
		Messages: []openai.ChatCompletionMessageParamUnion{
			openai.DeveloperMessage("Produce a concise answer to the query based on the provided sources."),
			openai.UserMessage(fmt.Sprintf("Sources: %s\n\nQuery: %q", formatResults(results.Data), userQuery)),
		},
	})
	if err != nil {
		panic(err)
	}
	fmt.Println(completion.Choices[0].Message.Content)
}

func formatResults(results []openai.VectorStoreSearchResponse) string {
	var sources strings.Builder
	sources.WriteString("<sources>")
	for _, result := range results {
		fmt.Fprintf(&sources, "<result file_id=%q file_name=%q>", result.FileID, result.Filename)
		for _, content := range result.Content {
			fmt.Fprintf(&sources, "<content>%s</content>", content.Text)
		}
		sources.WriteString("</result>")
	}
	sources.WriteString("</sources>")
	return sources.String()
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14
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
44import com.openai.client.OpenAIClient;
import com.openai.client.okhttp.OpenAIOkHttpClient;
import com.openai.models.chat.completions.ChatCompletionCreateParams;
import com.openai.models.vectorstores.VectorStoreSearchParams;
import java.util.stream.Collectors;

String vectorStoreId = "vs_123";

String query = "What is the return policy?";
var results =
    client
        .vectorStores()
        .search(vectorStoreId, VectorStoreSearchParams.builder().query(query).build());
String sources =
    results.data().stream()
        .map(
            result ->
                "<result file_id='"
                    + result.fileId()
                    + "' file_name='"
                    + result.filename()
                    + "'>"
                    + result.content().stream()
                        .map(content -> "<content>" + content.text() + "</content>")
                        .collect(Collectors.joining())
                    + "</result>")
        .collect(Collectors.joining());

var completion =
    client
        .chat()
        .completions()
        .create(
            ChatCompletionCreateParams.builder()
                .model("gpt-6-astra")
                .addDeveloperMessage(
                    "Answer the query concisely using only the provided sources.")
                .addUserMessage(
                    "Sources: <sources>" + sources + "</sources>\n\nQuery: " + query)
                .build());

completion.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
24require "openai"

client = OpenAI::Client.new
query = "What is the return policy?"
results = client.vector_stores.search("vs_123", query: query)
sources = (results.data || []).map do |result|
  content = result.content.map { |part| "<content>#{part.text}</content>" }.join
  "<result file_id='#{result.file_id}' file_name='#{result.filename}'>#{content}</result>"
end.join

completion = client.chat.completions.create(
  model: "gpt-6-astra",
  messages: [
    {
      role: :developer,
      content: "Answer the query concisely using only the provided sources."
    },
    {
      role: :user,
      content: "Sources: <sources>#{sources}</sources>\n\nQuery: #{query}"
    }
  ]
)
puts(completion.choices.fetch(0).message.content)

"Our return policy allows returns within 30 days of purchase."

這裡使用了範例函式 format_results，其實作方式可以如下：

結果格式化函式範例

Python

1
2
3
4
5
6
7
8
9
10
11function formatResults(results) {
  let formattedResults = "";
  for (const result of results.data) {
    let formattedResult = `<result file_id='${result.file_id}' file_name='${result.filename}'>`;
    for (const part of result.content) {
      formattedResult += `<content>${part.text}</content>`;
    }
    formattedResults += formattedResult + "</result>";
  }
  return `<sources>${formattedResults}</sources>`;
}

1
2
3
4
5
6
7
8
9
10def format_results(results):
    formatted_results = ""
    for result in results.data:
        formatted_result = (
            f"<result file_id='{result.file_id}' file_name='{result.file_name}'>"
        )
        for part in result.content:
            formatted_result += f"<content>{part.text}</content>"
        formatted_results += formatted_result + "</result>"
    return f"<sources>{formatted_results}</sources>"

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31package main

import (
	"fmt"
	"strings"

	"github.com/openai/openai-go/v3"
)

func main() {
	results := []openai.VectorStoreSearchResponse{{
		FileID:   "file-12345",
		Filename: "woodchuck_policy.txt",
		Content:  []openai.VectorStoreSearchResponseContent{{Text: "Each passenger may carry up to two woodchucks."}},
	}}
	fmt.Println(formatResults(results))
}

func formatResults(results []openai.VectorStoreSearchResponse) string {
	var sources strings.Builder
	sources.WriteString("<sources>")
	for _, result := range results {
		fmt.Fprintf(&sources, "<result file_id=%q file_name=%q>", result.FileID, result.Filename)
		for _, content := range result.Content {
			fmt.Fprintf(&sources, "<content>%s</content>", content.Text)
		}
		sources.WriteString("</result>")
	}
	sources.WriteString("</sources>")
	return sources.String()
}

1
2
3
4
5
6
7
8
9
10
11
12
13
14results = [
  {
    file_id: "file-12345",
    filename: "woodchuck_policy.txt",
    content: [{ text: "Each passenger may carry up to two woodchucks." }]
  }
]

sources = results.map do |result|
  content = result.fetch(:content).map { |part| "<content>#{part.fetch(:text)}</content>" }.join
  "<result file_id=\"#{result.fetch(:file_id)}\" file_name=\"#{result.fetch(:filename)}\">#{content}</result>"
end

puts("<sources>#{sources.join}</sources>")

建議

開始使用

核心概念

SDK 與 CLI

資源

舊版 API

選擇模型

文字與程式碼

提示詞

推理

圖像與影片

即時互動與音訊

專用模型

智慧體 API

智慧體 SDK

ChatKit

搜尋與檢索

連接工具與資料

建立工具工作流程

電腦與程式碼

媒體

GPT-Live

Realtime API

開發語音應用

連線

音訊處理

正式上線

效能與品質

成本與吞吐量

安全與治理

基礎架構與存取

營運

核心概念

Plan

Build

Test and publish

Conversion specs

指南

資源

開始使用

指南

File Upload

API

Measurement

Advertiser API

API 參考文件

開始使用

基礎

探索

適用平台

版本資訊

工作流程

能力

參考資料

自訂

設定檔

智慧體組態

擴充 ChatGPT 與 Codex 的功能

Linux

Windows

開發工作流程

擴充與自動化

環境

使用 Codex 進行建置

第三方整合

參考資料

權限

Codex Security

網路安全

開始使用

ChatGPT Work

身分識別與身分驗證

工作區存取權、政策與模型

外掛程式與連接器控制措施

使用情況、治理與合規

部署與模型供應商

社群

部落格

社群

部落格