Amazon 泰坦嵌入文本

Titan Embeddings G1 - Text不支援使用推論參數。以下各節詳細說明請求和回應格式，並提供程式碼範例。

請求和回應

請求主體在請InvokeModel求的body字段中傳遞。

V2 Request

輸入文字參數是必需的。標準化和標註參數是可選的。

• 輸入文字 — 輸入要轉換為嵌入的文字。

• 規範化-指示是否規範化輸出嵌入的標誌。預設為 true。

• 維度-輸出嵌入應具有的維度數。接受下列值：1024 (預設值)、512、256。

{
                        "inputText": string,
                        "dimensions": int,
                        "normalize": boolean
                        }

V2 Response

這些 欄位如下所述。

• 嵌入-array，代表您提供的輸入的嵌入向量。

• inputTextToken計數-輸入中令牌的數量。

{
    "embedding": [float, float, ...],
    "inputTextTokenCount": int
}

G1 Request

唯一可用的欄位是inputText，您可以在其中包含要轉換為嵌入的文字。

{
    "inputText": string
}

G1 Response

body的回應包含下列欄位。

{
    "embedding": [float, float, ...],
    "inputTextTokenCount": int
}

這些 欄位如下所述。

• 嵌入-array，代表您提供的輸入的嵌入向量。

• inputTextToken計數-輸入中令牌的數量。

範例程式碼

此範例顯示如何呼叫 Amazon Titan 嵌入模型來產生嵌入。

# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
# SPDX-License-Identifier: Apache-2.0
"""
Shows how to generate embeddings with the Amazon Titan Embeddings G1 - Text model (on demand).
"""

import json
import logging
import boto3


from botocore.exceptions import ClientError


logger = logging.getLogger(__name__)
logging.basicConfig(level=logging.INFO)


def generate_embeddings(model_id, body):
    """
    Generate a vector of embeddings for a text input using Amazon Titan Embeddings G1 - Text on demand.
    Args:
        model_id (str): The model ID to use.
        body (str) : The request body to use.
    Returns:
        response (JSON): The text that the model generated, token information, and the
        reason the model stopped generating text.
    """

    logger.info("Generating embeddings with Amazon Titan Embeddings G1 - Text model %s", model_id)

    bedrock = boto3.client(service_name='bedrock-runtime')

    accept = "application/json"
    content_type = "application/json"

    response = bedrock.invoke_model(
        body=body, modelId=model_id, accept=accept, contentType=content_type
    )

    response_body = json.loads(response.get('body').read())

    return response_body


def main():
    """
    Entrypoint for Amazon Titan Embeddings G1 - Text example.
    """

    logging.basicConfig(level=logging.INFO,
                        format="%(levelname)s: %(message)s")

    model_id = "amazon.titan-embed-text-v1"
    input_text = "What are the different services that you offer?"


    # Create request body.
    body = json.dumps({
        "inputText": input_text,
    })


    try:

        response = generate_embeddings(model_id, body)

        print(f"Generated embeddings: {response['embedding']}")
        print(f"Input Token count:  {response['inputTextTokenCount']}")

    except ClientError as err:
        message = err.response["Error"]["Message"]
        logger.error("A client error occurred: %s", message)
        print("A client error occured: " +
              format(message))

    else:
        print(f"Finished generating embeddings with Amazon Titan Embeddings G1 - Text model {model_id}.")


if __name__ == "__main__":
    main()

"""
Shows how to generate embeddings with the Amazon Titan Text Embeddings V2 Model
"""

import json
import logging
import boto3


from botocore.exceptions import ClientError


logger = logging.getLogger(__name__)
logging.basicConfig(level=logging.INFO)


def generate_embeddings(model_id, body):
    """
    Generate a vector of embeddings for a text input using Amazon Titan Text Embeddings G1 on demand.
    Args:
        model_id (str): The model ID to use.
        body (str) : The request body to use.
    Returns:
        response (JSON): The text that the model generated, token information, and the
        reason the model stopped generating text.
    """

    logger.info("Generating embeddings with Amazon Titan Text Embeddings V2 model %s", model_id)

    bedrock = boto3.client(service_name='bedrock-runtime')

    accept = "application/json"
    content_type = "application/json"

    response = bedrock.invoke_model(
        body=body, modelId=model_id, accept=accept, contentType=content_type
    )

    response_body = json.loads(response.get('body').read())

    return response_body


def main():
    """
    Entrypoint for Amazon Titan Embeddings V2 - Text example.
    """

    logging.basicConfig(level=logging.INFO,
                        format="%(levelname)s: %(message)s")

    model_id = "amazon.titan-embed-text-v2:0"
    input_text = "What are the different services that you offer?"


    # Create request body.
    body = json.dumps({
        "inputText": input_text,
        "dimensions": 512,
        "normalize": True
    })


    try:

        response = generate_embeddings(model_id, body)

        print(f"Generated embeddings: {response['embedding']}")
        print(f"Input Token count:  {response['inputTextTokenCount']}")

    except ClientError as err:
        message = err.response["Error"]["Message"]
        logger.error("A client error occurred: %s", message)
        print("A client error occured: " +
              format(message))

    else:
        print(f"Finished generating embeddings with Amazon Titan Text Embeddings V2 model {model_id}.")


if __name__ == "__main__":
    main()
        

隨時配置您的準確性成本權衡

雖然可以通過 API 獲得標準化，但客戶還可以在生成嵌入後減少嵌入維度，從而可以隨著需求的發展在準確性和成本之間進行折衷。這使客戶能夠產生 1024-dim 索引嵌入，將其存放在 S3 等低成本儲存選項中，並隨時將其 1024、512 或 256 維度版本載入他們喜愛的向量資料庫中。

若要將指定的內嵌從 1024 減少為 256 個維度，您可以使用下列範例邏輯：

import numpy as np
from numpy import linalg

def normalize_embedding(embedding: np.Array):
   '''
   Args:
       embedding: Unnormlized 1D/2D numpy array
              - 1D: (emb_dim)
             - 2D: (batch_size, emb_dim)
   Return:
       np.array: Normalized 1D/2D numpy array 
   '''
   return embedding/linalg.norm(embedding, dim=-1, keep_dim=True)
   
def reduce_emb_dim(embedding: np.Array, target_dim:int, normalize:bool=True) -> np.Array:
   '''
   Args:
       embedding: Unnormlized 1D/2D numpy array. Expected shape:
          - 1D: (emb_dim)
          - 2D: (batch_size, emb_dim)
       target_dim: target dimension to reduce the embedding to
   Return:
       np.array: Normalized 1D numpy array 
   '''
   smaller_embedding = embedding[..., :target_dim]
   if normalize:
      smaller_embedding = normalize_embedding(smaller_embedding)
   return smaller_embedding
   
   
if __name__ == '__main__':
   embedding = # bedrock client call
   reduced_embedding = # bedrock client call with dim=256
   post_reduction_embeddings = reduce_emb_dim(np.array(embeddings), dim=256)
   print(linalg.norm(np.array(reduced_embedding) - post_reduction_embeddings))