anthropic_client

Anthropic ModelClient integration.

Functions

get_first_message_content(completion)	When we only need the content of the first message.
handle_streaming_response(generator)	Handle the streaming response.

Classes

AnthropicAPIClient([api_key, ...])	A component wrapper for the Anthropic API client.
LLMResponse(*[, content, thinking, tool_use])	LLM Response

class AnthropicAPIClient(api_key: str | None = None, non_streaming_chat_completion_parser: Callable[[Completion], Any] = None, streaming_chat_completion_parser: Callable[[Completion], Any] = None)

Bases: ModelClient

A component wrapper for the Anthropic API client.

Visit https://docs.anthropic.com/en/docs/intro-to-claude for more api details.

Note:

As antropic API needs users to set max_tokens, we set up a default value of 512 for the max_tokens. You can override this value by passing the max_tokens in the model_kwargs.

Reference: - https://docs.anthropic.com/en/docs/about-claude/models - interlevad thinking:https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking#interleaved-thinking

init_sync_client()

init_async_client()

parse_chat_completion(completion: Completion | Generator[MessageStreamManager, None, None]) → GeneratorOutput

Parse the completion, and put it into the raw_response.

track_completion_usage(completion: Message) → CompletionUsage

Track the completion usage.

convert_inputs_to_api_kwargs(input: Any | None = None, model_kwargs: Dict = {}, model_type: ModelType = ModelType.UNDEFINED) → dict

Anthropic API messages separates the system and the user messages.

As we focus on one prompt, we have to use the user message as the input.

api: https://docs.anthropic.com/en/api/messages

call(api_kwargs: Dict = {}, model_type: ModelType = ModelType.UNDEFINED)

kwargs is the combined input and model_kwargs

async acall(api_kwargs: Dict = {}, model_type: ModelType = ModelType.UNDEFINED)

kwargs is the combined input and model_kwargs

get_first_message_content(completion: Message) → LLMResponse

When we only need the content of the first message. It is the default parser for chat completion.