View a markdown version of this page

Biblioteca de códigos - Amazon Nova

Biblioteca de códigos

Esta seção fornece exemplos de código para operações comuns do Amazon Nova usando a API Converse ou a API InvokeModel.

Exemplos da API Converse

Solicitação básica

Envie uma solicitação de texto básica para modelos do Amazon Nova usando a API Converse.

Non-streaming
import boto3 from botocore.config import Config # Create the Bedrock Runtime client, using an extended timeout configuration # to support long-running requests. bedrock = boto3.client( "bedrock-runtime", region_name="us-east-1", config=Config(read_timeout=3600), ) # Invoke the model response = bedrock.converse( modelId="us.amazon.nova-2-lite-v1:0", messages=[ { "role": "user", "content": [{"text": "Write a short story. End the story with 'THE END'."}], } ], system=[{"text": "You are a children's book author."}], # Optional inferenceConfig={ # These parameters are optional "maxTokens": 1500, "temperature": 0.7, "topP": 0.9, "stopSequences": ["THE END"], }, additionalModelRequestFields={ # These parameters are optional "inferenceConfig": { "topK": 50, } }, ) # Extract the text response content_list = response["output"]["message"]["content"] for content in content_list: if "text" in content: print(content["text"])
Streaming
import boto3 from botocore.config import Config # Create the Bedrock Runtime client, using an extended timeout configuration # to support long-running requests. bedrock = boto3.client( "bedrock-runtime", region_name="us-east-1", config=Config(connect_timeout=3600, read_timeout=3600), ) # Invoke the model response = bedrock.converse_stream( modelId="us.amazon.nova-2-lite-v1:0", messages=[ { "role": "user", "content": [{"text": "Write a short story. End the story with 'THE END'."}], } ], system=[{"text": "You are a children's book author."}], # Optional inferenceConfig={ # These parameters are optional "maxTokens": 1500, "temperature": 0.7, "topP": 0.9, "stopSequences": ["THE END"], }, additionalModelRequestFields={ # These parameters are optional "inferenceConfig": { "topK": 50, } }, ) # Handle streaming events for event in response["stream"]: if "contentBlockDelta" in event: delta = event["contentBlockDelta"]["delta"] if "text" in delta: print(delta["text"], end="", flush=True)

Entrada multimodal usando ativo incorporado

Processe conteúdo multimodal incorporando dados de documentos, imagens, vídeos ou áudios diretamente na solicitação. Este exemplo usa dados de imagem. Para obter detalhes sobre a estrutura de conteúdo para outras modalidades, consulte os detalhes do ContentBlock na documentação da API do Amazon Bedrock.

Non-streaming
import boto3 from botocore.config import Config # Read a document, image, video, or audio file with open("sample_image.png", "rb") as image_file: binary_data = image_file.read() data_format = "png" # Define message with image messages = [ { "role": "user", "content": [ { "image": { "format": data_format, "source": { "bytes": binary_data # For Invoke API, encode as Base64 string }, }, }, {"text": "Provide a brief caption for this asset."}, ], } ] # Create the Bedrock Runtime client, using an extended timeout configuration # to support long-running requests. bedrock = boto3.client( "bedrock-runtime", region_name="us-east-1", config=Config(read_timeout=3600), ) # Invoke model response = bedrock.converse( modelId="us.amazon.nova-2-lite-v1:0", messages=messages, ) # Extract the text response content_list = response["output"]["message"]["content"] for content in content_list: if "text" in content: print(content["text"])
Streaming
import boto3 from botocore.config import Config # Read a document, image, video, or audio file with open("sample_image.png", "rb") as image_file: binary_data = image_file.read() data_format = "png" # Define message with image messages = [ { "role": "user", "content": [ { "image": { "format": data_format, "source": { "bytes": binary_data # For Invoke API, encode as Base64 string }, }, }, {"text": "Provide a brief caption for this asset."}, ], } ] # Create the Bedrock Runtime client, using an extended timeout configuration # to support long-running requests. bedrock = boto3.client( "bedrock-runtime", region_name="us-east-1", config=Config(connect_timeout=3600, read_timeout=3600), ) # Invoke model with streaming response = bedrock.converse_stream( modelId="us.amazon.nova-2-lite-v1:0", messages=messages, ) # Handle streaming events for event in response["stream"]: if "contentBlockDelta" in event: delta = event["contentBlockDelta"]["delta"] if "text" in delta: print(delta["text"], end="", flush=True)

Entrada multimodal usando o URI do S3

Processe conteúdo multimodal referenciando documentos, imagens, vídeos ou arquivos de áudio armazenados no S3. Este exemplo usa uma referência de imagem. Para obter detalhes sobre a estrutura de conteúdo para outras modalidades, consulte os detalhes do ContentBlock na documentação da API do Amazon Bedrock.

Non-streaming
import boto3 from botocore.config import Config # Define message with image messages = [ { "role": "user", "content": [ { "image": { "format": "png", "source": { "s3Location": { "uri": "s3://path/to/your/asset", # "bucketOwner": "<account_id>" # Optional } }, }, }, {"text": "Provide a brief caption for this asset."}, ], } ] # Create the Bedrock Runtime client, using an extended timeout configuration # to support long-running requests. bedrock = boto3.client( "bedrock-runtime", region_name="us-east-1", config=Config(read_timeout=3600), ) # Invoke model response = bedrock.converse( modelId="us.amazon.nova-2-lite-v1:0", messages=messages, ) # Extract the text response content_list = response["output"]["message"]["content"] for content in content_list: if "text" in content: print(content["text"])
Streaming
import boto3 from botocore.config import Config # Define message with image messages = [ { "role": "user", "content": [ { "image": { "format": "png", "source": { "s3Location": { "uri": "s3://path/to/your/asset", # "bucketOwner": "<account_id>" # Optional } }, }, }, {"text": "Provide a brief caption for this asset."}, ], } ] # Create the Bedrock Runtime client, using an extended timeout configuration # to support long-running requests. bedrock = boto3.client( "bedrock-runtime", region_name="us-east-1", config=Config(connect_timeout=3600, read_timeout=3600), ) # Invoke model with streaming response = bedrock.converse_stream( modelId="us.amazon.nova-2-lite-v1:0", messages=messages, ) # Handle streaming events for event in response["stream"]: if "contentBlockDelta" in event: delta = event["contentBlockDelta"]["delta"] if "text" in delta: print(delta["text"], end="", flush=True)

Pensamento estendido (raciocínio)

Habilite o pensamento estendido para tarefas complexas de resolução de problemas.

Non-streaming
import boto3 from botocore.config import Config # Create the Bedrock Runtime client, using an extended timeout configuration # to support long-running requests. bedrock = boto3.client( "bedrock-runtime", region_name="us-east-1", config=Config(read_timeout=3600), ) # Invoke the model response = bedrock.converse( modelId="us.amazon.nova-2-lite-v1:0", messages=[ { "role": "user", "content": [ { "text": 'How many capital letters appear in the following passage. Your response must include only the number: "Wilfred ordered an anvil from ACME. Shipping was expensive."' } ], } ], additionalModelRequestFields={ "reasoningConfig": { "type": "enabled", "maxReasoningEffort": "low", # "low" | "medium" | "high" } }, ) # Extract response content content_list = response["output"]["message"]["content"] for content in content_list: # Extract the reasoning response if "reasoningContent" in content: print("\n== Reasoning ==") print(content["reasoningContent"]["reasoningText"]["text"]) # Extract the text response if "text" in content: print("\n== Text ==") print(content["text"])
Streaming
import boto3 from botocore.config import Config # Create the Bedrock Runtime client, using an extended timeout configuration # to support long-running requests. bedrock = boto3.client( "bedrock-runtime", region_name="us-east-1", config=Config(connect_timeout=3600, read_timeout=3600), ) # Invoke the model response = bedrock.converse_stream( modelId="us.amazon.nova-2-lite-v1:0", messages=[ { "role": "user", "content": [ { "text": 'How many capital letters appear in the following passage. Your response must include only the number: "Wilfred ordered an anvil from ACME. Shipping was expensive."' } ], } ], additionalModelRequestFields={ "reasoningConfig": { "type": "enabled", "maxReasoningEffort": "low", # "low" | "medium" | "high" }, }, ) # Process the streaming response reasoning_output = "" text_output = "" for event in response["stream"]: if "contentBlockDelta" in event: delta = event["contentBlockDelta"]["delta"] if "reasoningContent" in delta: if len(reasoning_output) == 0: print("\n\n== Reasoning ==") reasoning_text_chunk = delta["reasoningContent"]["text"] print(reasoning_text_chunk, end="", flush=True) reasoning_output += reasoning_text_chunk elif "text" in delta: if len(text_output) == 0: print("\n\n== Text ==") text_chunk = delta["text"] print(text_chunk, end="", flush=True) text_output += text_chunk

Ferramenta integrada: Ancoragem do Nova com citações

Use a Ancoragem do Nova para recuperar informações em tempo real da web com citações.

Non-streaming
import boto3 from botocore.config import Config # Define the list of tools the model may use tool_config = {"tools": [{"systemTool": {"name": "nova_grounding"}}]} # Create the Bedrock Runtime client, using an extended timeout configuration # to support long-running requests. bedrock = boto3.client( "bedrock-runtime", region_name="us-east-1", config=Config(read_timeout=3600), ) messages = [ { "role": "user", "content": [ {"text": "What is the latest news about renewable energy sources?"} ], } ] # Invoke the model response = bedrock.converse( modelId="us.amazon.nova-2-lite-v1:0", messages=messages, toolConfig=tool_config ) # Extract the text with interleaved citations output_with_citations = "" content_list = response["output"]["message"]["content"] for content in content_list: if "text" in content: output_with_citations += content["text"] elif "citationsContent" in content: citations = content["citationsContent"]["citations"] for citation in citations: url = citation["location"]["web"]["url"] output_with_citations += f"[{url}]" print(output_with_citations)
Streaming
import boto3 from botocore.config import Config # Define the list of tools the model may use tool_config = {"tools": [{"systemTool": {"name": "nova_grounding"}}]} # Create the Bedrock Runtime client, using an extended timeout configuration # to support long-running requests. bedrock = boto3.client( "bedrock-runtime", region_name="us-east-1", config=Config(read_timeout=3600), ) messages = [ { "role": "user", "content": [ {"text": "What is the latest news about renewable energy sources?"} ], } ] # Invoke the model with streaming response = bedrock.converse_stream( modelId="us.amazon.nova-2-lite-v1:0", messages=messages, toolConfig=tool_config ) # Process the streaming response with interleaved citations for event in response["stream"]: if "contentBlockDelta" in event: delta = event["contentBlockDelta"]["delta"] if "text" in delta: print(delta["text"], end="", flush=True) elif "citation" in delta: url = delta["citation"]["location"]["web"]["url"] print(f"[{url}]", end="", flush=True)

Ferramenta integrada: Interpretador de Código

Use a ferramenta Interpretador de Código para executar código Python para cálculos e análise de dados.

Non-streaming
import boto3 from botocore.config import Config # Define the list of tools the model may use tool_config = {"tools": [{"systemTool": {"name": "nova_code_interpreter"}}]} # Create the Bedrock Runtime client, using an extended timeout configuration # to support long-running requests. bedrock = boto3.client( "bedrock-runtime", region_name="us-east-1", config=Config(read_timeout=3600), ) messages = [ { "role": "user", "content": [ { "text": "What is the average of 10, 24, 2, 3, 43, 52, 13, 68, 6, 7, 902, 82?" } ], } ] # Invoke the model response = bedrock.converse( modelId="us.amazon.nova-2-lite-v1:0", messages=messages, toolConfig=tool_config ) # Extract the text and the code the was executed content_list = response["output"]["message"]["content"] for content in content_list: if "text" in content: print("\n== Text ==") print(content["text"]) elif "toolUse" in content and content["toolUse"]["name"] == "nova_code_interpreter": print("\n== Code Interpreter: input.snippet ==") print(content["toolUse"]["input"]["snippet"])
Streaming
import boto3 from botocore.config import Config import json # Define the list of tools the model may use tool_config = {"tools": [{"systemTool": {"name": "nova_code_interpreter"}}]} messages = [ { "role": "user", "content": [ { "text": "What is the average of 10, 24, 2, 3, 43, 52, 13, 68, 6, 7, 902, 82?" } ], } ] # Create the Bedrock Runtime client, using an extended timeout configuration # to support long-running requests. bedrock = boto3.client( "bedrock-runtime", region_name="us-east-1", config=Config(connect_timeout=3600, read_timeout=3600), ) # Invoke the model with streaming response = bedrock.converse_stream( modelId="us.amazon.nova-2-lite-v1:0", messages=messages, toolConfig=tool_config ) # Process the streaming response current_block_start = None response_text = "" for event in response["stream"]: if "contentBlockStart" in event: current_block_start = event["contentBlockStart"]["start"] elif "contentBlockStop" in event: current_block_start = None elif "contentBlockDelta" in event: delta = event["contentBlockDelta"]["delta"] if ( current_block_start and "toolUse" in current_block_start and current_block_start["toolUse"]["name"] == "nova_code_interpreter" ): # This is code interpreter content tool_input = json.loads(delta["toolUse"]["input"]) print("\n== Executed Code Snippet ==") print(tool_input["snippet"], end="", flush=True) elif "text" in delta: # This is text response content if len(response_text) == 0: print("\n== Text ==") text = delta["text"] response_text += text print(text, end="", flush=True)

Uso de ferramentas

Defina ferramentas personalizadas para o modelo usar durante a conversa.

Non-streaming
import boto3 from botocore.config import Config def get_weather(city): # Mock function to simulate weather API return {"temperatureF": 48, "conditions": "light rain"} # Define the toolSpec for the weather tool weather_tool = { "toolSpec": { "name": "get_weather", "description": "Get the current weather conditions in a given location", "inputSchema": { "json": { "type": "object", "properties": { "city": { "type": "string", "description": "The city and state, e.g. San Francisco, CA", } }, "required": ["city"], } }, } } # Define the list of tools the model may use tool_config = {"tools": [weather_tool]} # Create the Bedrock Runtime client, using an extended timeout configuration # to support long-running requests. bedrock = boto3.client( "bedrock-runtime", region_name="us-east-1", config=Config(read_timeout=3600), ) # Start tracking message history messages = [] messages.append( { "role": "user", "content": [ { "text": "Suggest some activities to do in Seattle based on the current weather." } ], } ) # Invoke the model response = bedrock.converse( modelId="us.amazon.nova-2-lite-v1:0", messages=messages, toolConfig=tool_config ) assistant_message = response["output"]["message"] # Add the assistant response to the message history messages.append(assistant_message) content_list = assistant_message["content"] stop_reason = response["stopReason"] if stop_reason == "tool_use": # Extract the toolUse details tool_use = next( content["toolUse"] for content in content_list if "toolUse" in content ) tool_name = tool_use["name"] tool_use_id = tool_use["toolUseId"] if tool_name == "get_weather": # Call the tool weather = get_weather(tool_use["input"]["city"]) # Send the result back to the model messages.append( { "role": "user", "content": [ { "toolResult": { "toolUseId": tool_use_id, "content": [{"json": weather}], } } ], } ) # Submit the tool result back to the model response = bedrock.converse( modelId="us.amazon.nova-2-lite-v1:0", messages=messages, toolConfig=tool_config, ) content_list = response["output"]["message"]["content"] for content in content_list: # Extract the text response if "text" in content: print("\n== Text ==") print(content["text"]) else: # A tool call was not needed for content in content_list: # Extract the text response if "text" in content: print("\n== Text ==") print(content["text"])
Streaming
import boto3 from botocore.config import Config import json def get_weather(city): # Mock function to simulate weather API return {"temperatureF": 48, "conditions": "light rain"} # Define the toolSpec for the weather tool weather_tool = { "toolSpec": { "name": "get_weather", "description": "Get the current weather conditions in a given location", "inputSchema": { "json": { "type": "object", "properties": { "city": { "type": "string", "description": "The city and state, e.g. San Francisco, CA", } }, "required": ["city"], } }, } } # Define the list of tools the model may use tool_config = {"tools": [weather_tool]} # Create the Bedrock Runtime client, using an extended timeout configuration # to support long-running requests. bedrock = boto3.client( "bedrock-runtime", region_name="us-east-1", config=Config(read_timeout=3600), ) # Start tracking message history messages = [] messages.append( { "role": "user", "content": [ { "text": "Suggest some activities to do in Seattle based on the current weather." } ], } ) # Invoke the model with streaming response = bedrock.converse_stream( modelId="us.amazon.nova-2-lite-v1:0", messages=messages, toolConfig=tool_config ) # Process the streaming response assistant_message = {"role": "assistant", "content": []} current_tool_use = None stop_reason = None for event in response["stream"]: if "contentBlockStart" in event: start = event["contentBlockStart"]["start"] if "toolUse" in start: current_tool_use = start["toolUse"] current_tool_use["input"] = "" elif "contentBlockDelta" in event: delta = event["contentBlockDelta"]["delta"] if "toolUse" in delta: current_tool_use["input"] += delta["toolUse"]["input"] elif "text" in delta: print(delta["text"], end="", flush=True) elif "contentBlockStop" in event: if current_tool_use: # Parse the accumulated tool input current_tool_use["input"] = json.loads(current_tool_use["input"]) assistant_message["content"].append({"toolUse": current_tool_use}) current_tool_use = None elif "messageStop" in event: stop_reason = event["messageStop"]["stopReason"] if stop_reason == "end_turn": exit # Add the assistant response to the message history messages.append(assistant_message) if stop_reason == "tool_use": # Extract the toolUse details tool_use = next( content["toolUse"] for content in assistant_message["content"] if "toolUse" in content ) tool_name = tool_use["name"] tool_use_id = tool_use["toolUseId"] if tool_name == "get_weather": # Call the tool weather = get_weather(tool_use["input"]["city"]) # Send the result back to the model messages.append( { "role": "user", "content": [ { "toolResult": { "toolUseId": tool_use_id, "content": [{"json": weather}], } } ], } ) # Submit the tool result back to the model with streaming response = bedrock.converse_stream( modelId="us.amazon.nova-2-lite-v1:0", messages=messages, toolConfig=tool_config, ) # Handle the final streaming response print("\n== Text ==") for event in response["stream"]: if "contentBlockDelta" in event: delta = event["contentBlockDelta"]["delta"] if "text" in delta: print(delta["text"], end="", flush=True)

Exemplos da API InvokeModel

Os exemplos abaixo focam as poucas áreas-chave em que as estruturas de solicitação e resposta da API Invoke diferem levemente das da API Converse. Em quase todos os outros aspectos, as duas APIs são compatíveis, então você poderá adaptar facilmente os exemplos da API Converse acima para trabalhar com a API InvokeModel.

Solicitação básica

Envie uma solicitação de texto básica para os modelos do Amazon Nova 2 usando a API InvokeModel.

Non-streaming
import json import boto3 from botocore.config import Config # Configure the request request_body = { "messages": [ { "role": "user", "content": [{"text": "Write a short story. End the story with 'THE END'."}], } ], "system": [{"text": "You are a children's book author."}], # Optional "inferenceConfig": { # These parameters are optional "maxTokens": 1500, "temperature": 0.7, "topP": 0.9, "topK": 50, "stopSequences": ["THE END"], }, } bedrock = boto3.client( "bedrock-runtime", region_name="us-east-1", config=Config(read_timeout=3600), ) # Invoke the model response = bedrock.invoke_model( modelId="us.amazon.nova-2-lite-v1:0", body=json.dumps(request_body) ) response_body = json.loads(response["body"].read()) # Extract the text response content_list = response_body["output"]["message"]["content"] for content in content_list: if "text" in content: print(content["text"])
Streaming
import json import boto3 from botocore.config import Config # Configure the request request_body = { "messages": [ { "role": "user", "content": [{"text": "Write a short story. End the story with 'THE END'."}], } ], "system": [{"text": "You are a children's book author."}], # Optional "inferenceConfig": { # These parameters are optional "maxTokens": 1500, "temperature": 0.7, "topP": 0.9, "topK": 50, "stopSequences": ["THE END"], }, } bedrock = boto3.client( "bedrock-runtime", region_name="us-east-1", config=Config(connect_timeout=3600, read_timeout=3600), ) # Invoke the model with streaming response = bedrock.invoke_model_with_response_stream( modelId="us.amazon.nova-2-lite-v1:0", body=json.dumps(request_body) ) # Process the streaming response for event in response["body"]: chunk = json.loads(event["chunk"]["bytes"]) if "contentBlockDelta" in chunk: delta = chunk["contentBlockDelta"]["delta"] if "text" in delta: print(delta["text"], end="", flush=True)

API InvokeModel com raciocínio

Use a API InvokeModel com o raciocínio habilitado para a resolução de problemas complexos.

Non-streaming
import json import boto3 from botocore.config import Config # Configure the request request_body = { "messages": [ { "role": "user", "content": [ { "text": 'How many capital letters appear in the following passage. Your response must include only the number: "Wilfred ordered an anvil from ACME. Shipping was expensive."' } ], } ], "reasoningConfig": { "type": "enabled", "maxReasoningEffort": "low", # "low" | "medium" | "high" }, } bedrock = boto3.client( "bedrock-runtime", region_name="us-east-1", config=Config(read_timeout=3600), ) # Invoke the model response = bedrock.invoke_model( modelId="us.amazon.nova-2-lite-v1:0", body=json.dumps(request_body) ) response_body = json.loads(response["body"].read()) # Extract response content content_list = response_body["output"]["message"]["content"] for content in content_list: # Extract the reasoning response if "reasoningContent" in content: print("\n== Reasoning ==") print(content["reasoningContent"]["reasoningText"]["text"]) # Extract the text response if "text" in content: print("\n== Text ==") print(content["text"])
Streaming
import json import boto3 from botocore.config import Config # Configure the request request_body = { "messages": [ { "role": "user", "content": [ { "text": 'How many capital letters appear in the following passage. Your response must include only the number: "Wilfred ordered an anvil from ACME. Shipping was expensive."' } ], } ], "reasoningConfig": { "type": "enabled", "maxReasoningEffort": "low", # "low" | "medium" | "high" }, } bedrock = boto3.client( "bedrock-runtime", region_name="us-east-1", config=Config(connect_timeout=3600, read_timeout=3600), ) # Invoke the model with streaming response = bedrock.invoke_model_with_response_stream( modelId="us.amazon.nova-2-lite-v1:0", body=json.dumps(request_body) ) # Process the streaming response for event in response["body"]: chunk = json.loads(event["chunk"]["bytes"]) if "contentBlockDelta" in chunk: delta = chunk["contentBlockDelta"]["delta"] # Extract the reasoning response if "reasoningContent" in delta: print("\n== Reasoning ==") print(delta["reasoningContent"]["reasoningText"]["text"], end="", flush=True) # Extract the text response if "text" in delta: print("\n== Text ==") print(delta["text"], end="", flush=True)