langchain-cohere

Implements the Embeddings interface with Cohere's text representation language models.

Find out more about us at https://cohere.com and https://huggingface.co/CohereForAI

This implementation uses

Base class for Cohere models.

Implements the BaseChatModel (and BaseLanguageModel) interface with Cohere's large language models.

Find out more about us at https://cohere.com and https://huggingface.co/CohereForAI

This imple

Document compressor that uses Cohere Rerank API.

Parses a message into agent actions/finish.

Cut off the text as soon as any stop words occur.

Use tenacity to retry the completion call.

Use tenacity to retry the completion call.

Get the role of the message.

Get the role of the message (V2). Args: message: The message. Returns: The role of the message. Raises: ValueError: If the message is of an unknown type.

Get the request for the Cohere chat API (V2). Args: messages: The messages. **kwargs: The keyword arguments. Returns: The request for the Cohere chat API.

Renders the structured preamble part of the prompt content.

Renders a tool into prompt content. Either a BaseTool instance, or, a JSON schema must be provided.

Renders the 'output' part of an Agent's intermediate step into prompt content.

Converts observations into a 'document' dict

Renders an agent's intermediate steps into prompt content.

The returned function produces a BasePromptTemplate suitable for multi-hop.

directly_answer is a special tool that's always presented to the model as an available tool. The model only ever invokes this whilst answering and no AgentAction is produced, so it only needs to b

Renders the role of a message into prompt content.

Renders one or more BaseMessage implementations into prompt content.

Parses model-generated jsonified actions.

Expects input of the form
"{tool_use_prefix}: ```json

[{list of jsonified objects}]```"

outputs parsed list of jsonified objects.

Parses string into key-value pairs, according to patterns supplied in prefixes.

Also strips.

If inputs are:

completion = "
o: sam
bye then: paul.",
prefixes =

Parse action selections from model output.

Parses a grounded_generation (from parse_actions) and documents (from convert_to_documents) into a (generation, CohereCitation list) tuple.