探索AmazonBedrock的功能与应用

# 探索 Amazon Bedrock 的功能与应用 ## 1. 流式响应 流式响应对于响应式聊天界面特别有用，可维持交互式应用程序的实时性。以下是使用 Bedrock 的 InvokeModelWithResponseStream API 的 Python 代码示例： ```python response = bedrock_runtime.invoke_model_with_response_stream( modelId=modelId, body=body) stream = response.get('body') if stream: for event in stream: chunk=event.get('chunk') if chunk: print(json.loads(chunk.get('bytes').decode)) ``` ## 2. 大语言模型推理 API 基础模型会公开一组生成配置参数，这些参数会在推理过程中影响模型的输出。这些配置参数能让你控制模型的响应，包括令牌的多样性和数量。大多数模型支持温度（temperature）、top_k 和 top_p。 以下是使用 `invoke_model()` API 的 Bedrock 推理 API 请求示例，其中包含了使用 Bedrock 模型的提示配置参数： ```python import boto3 import json bedrock_runtime = boto3.client( service_name='bedrock-runtime' ) prompt = "<your prompt here>" body = json.dumps({ "inputText": "This is where you place your input text", "textGenerationConfig": { "temperature":0, "topP":1 } }) modelId = '...' # Amazon Bedrock 基础模型: # Amazon Titan Text # Anthropic Claude # AI21 Jurassic # Cohere Command # Meta Llama2 # 等等 response = bedrock_runtime.invoke_model( body=body, modelId=modelId) response_body = json.loads(response.get('body').read()) print(response_body.get('results')[0].get('outputText')) ``` ## 3. 生成 SQL 代码 许多文本生成模型（包括 Amazon Bedrock 中的模型）都在大量文本数据（包括代码示例）上进行了预训练。以下是使用 Amazon Bedrock 生成 SQL 查询的示例： ```python prompt = """ I have a table called 'students' with fields 'id', 'age', 'year_enrollment', 'subject', 'grade'. Write me a SQL Query that returns the 'id' with the highest 'age'. """ body = json.dumps({"inputText": prompt}) modelId = '...' response = bedrock_runtime.invoke_model( body=body, modelId=modelId) response_body = json.loads(response.get('body').read()) print(response_body.get('results')[0].get('outputText')) ``` ## 4. 文本摘要 生成文本摘要也是生成式 AI 的一个流行用例。以下是构建提示并发送 API 请求的步骤： 1. 构建提示： ```python prompt = """ Please provide a summary of the following text. Do not add any information that is not mentioned in the text below. <text> AWS took all of that feedback from customers, and today we are excited to announce Amazon Bedrock, a new service that makes generative foundation models accessible via an API. Bedrock is the easiest way for customers to build and scale generative AI-based applications using FMs, democratizing access for all builders. </text> """ ``` 2. 定义 API 请求体： ```python body = json.dumps( { "inputText": prompt, "textGenerationConfig":{ "maxTokenCount":128, "temperature":0, "topP":1 } } ) ``` 3. 发送 API 请求： ```python import json response = bedrock_runtime.invoke_model_with_response_stream( body=body, modelId=modelId) stream = response.get('body') output = [] if stream: for event in stream: chunk = event.get('chunk') if chunk: chunk_obj = json.loads(chunk.get('bytes').decode()) text = chunk_obj['outputText'] output.append(text) print(''.join(output)) ``` ## 5. 嵌入生成 嵌入是生成式 AI 和机器学习中的关键概念。可以使用 Amazon Bedrock 模型为任何输入字符串检索嵌入向量，然后比较向量之间的距离以找到最相关的文本字符串。常见用例包括语义搜索、推荐和分类。 以下是生成嵌入向量的代码示例： ```python def get_embedding(body, modelId, accept, contentType): response = bedrock_runtime.invoke_model( body=body, modelId=modelId) response_body = json.loads(response.get('body').read()) embedding = response_body.get('embedding') return embedding body = json.dumps( { "inputText": "<your prompt here>" } ) modelId = '...' embedding = get_embedding(body, modelId) print(embedding) ``` 为了可视化句子之间的语义相似性，可以生成热力图： ```python import sklearn from sklearn.preprocessing import normalize import numpy as np import seaborn as sns def plot_similarity_heatmap(text_labels, embeddings, rotation): inner_product = np.inner(embeddings, embeddings) sns.set(font_scale=1.1) graph = sns.heatmap( inner_product, xticklabels=text_labels, yticklabels=text_labels, vmin=np.min(inner_product), vmax=1, cmap="BuPu", ) graph.set_xticklabels(text_labels, rotation=rotation) graph.set_title("Semantic Textual Similarity Between Sentences") phrases = [ # Animals "Shelbee's dog, Molly, is so cute.", "Antje hates cats.", "Chris's first dog was very cute.", # U.S. Cities "Chicago is the place where I'm from.", "I work in San Francisco.", "Washington D.C. is a great place to visit.", # Color "What is your favorite color?", "Is Molly brown?", "Are your eyes blue?" ] embeddings = [] for phrase in phrases: query_response = get_embedding( body=json.dumps({"inputText": phrase}), modelId="...") embeddings.append(query_response) # 内积之前进行归一化 embeddings = normalize(np.array(embeddings), axis=1) %matplotlib inline %config InlineBackend.figure_format = 'retina' plot_similarity_heatmap(phrases, embeddings, 90) ``` ## 6. 模型微调 当你决定自定义模型时，可以使用 Amazon Bedrock 对基础模型进行微调。以下是微调模型的步骤： 1. 准备数据集：将数据集以 JSON Lines 格式存储在 S3 中，例如： ```json {"prompt": "I love going to the movies", "completion": "Positive"} {"prompt": "This new shirt is gorgeous", "completion": "Positive"} {"prompt": "The weather is awful", "completion": "Negative"} {"prompt": "This movie is terrible", "completion": "Negative"} ``` 2. 调用 `create_model_customization_job()` 开始微调： ```python import boto3 bedrock = boto3.client(service_name='bedrock') input_training_data = "s3://<BUCKET>/train.jsonl" output_data = "s3://<BUCKET>/output/" bedrock.create_model_customization_job( jobName="my-job", customModelName="my-fine-tuned-model", baseModelIdentifier="...", # Bedrock 基础模型 trainingDataConfig={"s3Uri": input_training_data}, outputDataConfig={"s3Uri": output_data}, ```