202 lines
8.1 KiB
Python
202 lines
8.1 KiB
Python
import os
|
|
import json
|
|
from typing import Dict, List, Optional, Any
|
|
|
|
import boto3
|
|
|
|
from mem0.llms.base import LLMBase
|
|
from mem0.configs.llms.base import BaseLlmConfig
|
|
|
|
class AWSBedrockLLM(LLMBase):
|
|
def __init__(self, config: Optional[BaseLlmConfig] = None):
|
|
super().__init__(config)
|
|
|
|
if not self.config.model:
|
|
self.config.model="anthropic.claude-3-5-sonnet-20240620-v1:0"
|
|
self.client = boto3.client("bedrock-runtime", region_name=os.environ.get("AWS_REGION"), aws_access_key_id=os.environ.get("AWS_ACCESS_KEY"), aws_secret_access_key=os.environ.get("AWS_SECRET_ACCESS_KEY"))
|
|
self.model_kwargs = {"temperature": self.config.temperature, "max_tokens_to_sample": self.config.max_tokens, "top_p": self.config.top_p}
|
|
|
|
def _format_messages(self, messages: List[Dict[str, str]]) -> str:
|
|
"""
|
|
Formats a list of messages into the required prompt structure for the model.
|
|
|
|
Args:
|
|
messages (List[Dict[str, str]]): A list of dictionaries where each dictionary represents a message.
|
|
Each dictionary contains 'role' and 'content' keys.
|
|
|
|
Returns:
|
|
str: A formatted string combining all messages, structured with roles capitalized and separated by newlines.
|
|
"""
|
|
formatted_messages = []
|
|
for message in messages:
|
|
role = message['role'].capitalize()
|
|
content = message['content']
|
|
formatted_messages.append(f"\n\n{role}: {content}")
|
|
|
|
return "".join(formatted_messages) + "\n\nAssistant:"
|
|
|
|
def _parse_response(self, response, tools) -> str:
|
|
"""
|
|
Process the response based on whether tools are used or not.
|
|
|
|
Args:
|
|
response: The raw response from API.
|
|
tools: The list of tools provided in the request.
|
|
|
|
Returns:
|
|
str or dict: The processed response.
|
|
"""
|
|
if tools:
|
|
processed_response = {
|
|
"tool_calls": []
|
|
}
|
|
|
|
if response["output"]["message"]["content"]:
|
|
for item in response["output"]["message"]["content"]:
|
|
if "toolUse" in item:
|
|
processed_response["tool_calls"].append({
|
|
"name": item["toolUse"]["name"],
|
|
"arguments": item["toolUse"]["input"]
|
|
})
|
|
|
|
return processed_response
|
|
|
|
response_body = json.loads(response['body'].read().decode())
|
|
return response_body.get('completion', '')
|
|
|
|
def _prepare_input(
|
|
self,
|
|
provider: str,
|
|
model: str,
|
|
prompt: str,
|
|
model_kwargs: Optional[Dict[str, Any]] = {},
|
|
) -> Dict[str, Any]:
|
|
"""
|
|
Prepares the input dictionary for the specified provider's model by mapping and renaming
|
|
keys in the input based on the provider's requirements.
|
|
|
|
Args:
|
|
provider (str): The name of the service provider (e.g., "meta", "ai21", "mistral", "cohere", "amazon").
|
|
model (str): The name or identifier of the model being used.
|
|
prompt (str): The text prompt to be processed by the model.
|
|
model_kwargs (Dict[str, Any]): Additional keyword arguments specific to the model's requirements.
|
|
|
|
Returns:
|
|
Dict[str, Any]: The prepared input dictionary with the correct keys and values for the specified provider.
|
|
"""
|
|
|
|
input_body = {"prompt": prompt, **model_kwargs}
|
|
|
|
provider_mappings = {
|
|
"meta": {"max_tokens_to_sample": "max_gen_len"},
|
|
"ai21": {"max_tokens_to_sample": "maxTokens", "top_p": "topP"},
|
|
"mistral": {"max_tokens_to_sample": "max_tokens"},
|
|
"cohere": {"max_tokens_to_sample": "max_tokens", "top_p": "p"},
|
|
}
|
|
|
|
if provider in provider_mappings:
|
|
for old_key, new_key in provider_mappings[provider].items():
|
|
if old_key in input_body:
|
|
input_body[new_key] = input_body.pop(old_key)
|
|
|
|
if provider == "cohere" and "cohere.command-r" in model:
|
|
input_body["message"] = input_body.pop("prompt")
|
|
|
|
if provider == "amazon":
|
|
input_body = {
|
|
"inputText": prompt,
|
|
"textGenerationConfig": {
|
|
"maxTokenCount": model_kwargs.get("max_tokens_to_sample"),
|
|
"topP": model_kwargs.get("top_p"),
|
|
"temperature": model_kwargs.get("temperature")
|
|
}
|
|
}
|
|
input_body["textGenerationConfig"] = {k: v for k, v in input_body["textGenerationConfig"].items() if v is not None}
|
|
|
|
return input_body
|
|
|
|
def _convert_tool_format(self, original_tools):
|
|
"""
|
|
Converts a list of tools from their original format to a new standardized format.
|
|
|
|
Args:
|
|
original_tools (list): A list of dictionaries representing the original tools, each containing a 'type' key and corresponding details.
|
|
|
|
Returns:
|
|
list: A list of dictionaries representing the tools in the new standardized format.
|
|
"""
|
|
new_tools = []
|
|
|
|
for tool in original_tools:
|
|
if tool['type'] == 'function':
|
|
function = tool['function']
|
|
new_tool = {
|
|
"toolSpec": {
|
|
"name": function['name'],
|
|
"description": function['description'],
|
|
"inputSchema": {
|
|
"json": {
|
|
"type": "object",
|
|
"properties": {},
|
|
"required": function['parameters'].get('required', [])
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
for prop, details in function['parameters'].get('properties', {}).items():
|
|
new_tool["toolSpec"]["inputSchema"]["json"]["properties"][prop] = {
|
|
"type": details.get('type', 'string'),
|
|
"description": details.get('description', '')
|
|
}
|
|
|
|
new_tools.append(new_tool)
|
|
|
|
return new_tools
|
|
|
|
def generate_response(
|
|
self,
|
|
messages: List[Dict[str, str]],
|
|
tools: Optional[List[Dict]] = None,
|
|
tool_choice: str = "auto",
|
|
):
|
|
"""
|
|
Generate a response based on the given messages using AWS Bedrock.
|
|
|
|
Args:
|
|
messages (list): List of message dicts containing 'role' and 'content'.
|
|
tools (list, optional): List of tools that the model can call. Defaults to None.
|
|
tool_choice (str, optional): Tool choice method. Defaults to "auto".
|
|
|
|
Returns:
|
|
str: The generated response.
|
|
"""
|
|
|
|
if tools:
|
|
# Use converse method when tools are provided
|
|
messages = [{"role": "user", "content": [{"text": message["content"]} for message in messages]}]
|
|
inference_config = {"temperature": self.model_kwargs["temperature"], "maxTokens": self.model_kwargs["max_tokens_to_sample"], "topP": self.model_kwargs["top_p"]}
|
|
tools_config = {"tools": self._convert_tool_format(tools)}
|
|
|
|
response = self.client.converse(
|
|
modelId=self.config.model,
|
|
messages=messages,
|
|
inferenceConfig=inference_config,
|
|
toolConfig=tools_config
|
|
)
|
|
else:
|
|
# Use invoke_model method when no tools are provided
|
|
prompt = self._format_messages(messages)
|
|
provider = self.model.split(".")[0]
|
|
input_body = self._prepare_input(provider, self.config.model, prompt, **self.model_kwargs)
|
|
body = json.dumps(input_body)
|
|
|
|
response = self.client.invoke_model(
|
|
body=body,
|
|
modelId=self.model,
|
|
accept='application/json',
|
|
contentType='application/json'
|
|
)
|
|
|
|
return self._parse_response(response, tools)
|