Marc•13mo ago

more on tags metadata

>> more on tags/metadata

21 Replies

Marc•13mo ago

@Luke Skywalker what's your use case for metadata? would tags or metadata on chains/agents be helpful? From Langchain Base Chain

tags: Optional[List[str]] = None
    """Optional list of tags associated with the chain. Defaults to None.
    These tags will be associated with each call to this chain,
    and passed as arguments to the handlers defined in `callbacks`.
    You can use these to eg identify a specific instance of a chain with its use case.
    """
    metadata: Optional[Dict[str, Any]] = None
    """Optional metadata associated with the chain. Defaults to None.
    This metadata will be associated with each call to this chain,
    and passed as arguments to the handlers defined in `callbacks`.
    You can use these to eg identify a specific instance of a chain with its use case.
    """

tags: Optional[List[str]] = None
    """Optional list of tags associated with the chain. Defaults to None.
    These tags will be associated with each call to this chain,
    and passed as arguments to the handlers defined in `callbacks`.
    You can use these to eg identify a specific instance of a chain with its use case.
    """
    metadata: Optional[Dict[str, Any]] = None
    """Optional metadata associated with the chain. Defaults to None.
    This metadata will be associated with each call to this chain,
    and passed as arguments to the handlers defined in `callbacks`.
    You can use these to eg identify a specific instance of a chain with its use case.
    """

Luke Skywalker•13mo ago

Use case is person identification and personal data extraction from document datasets. I put tags on chain already.

extracted_datapoint = extract_datapoint_chain.run(
                        query_extract,
                        tags=[
                            extract_model,
                            "extract_data_default",
                            retrieved_metadatas_filtered[single_doc_index]["title"],
                            f"relevance_distance={distance_array_filtered[single_doc_index]}",
                            # single_doc["metadata"]["title"],
                            property_name,
                            f"{firstname} {lastname}'s {property_details['title']}",
                        ],
                        callbacks=[langfuse_handler],
                    )

extracted_datapoint = extract_datapoint_chain.run(
                        query_extract,
                        tags=[
                            extract_model,
                            "extract_data_default",
                            retrieved_metadatas_filtered[single_doc_index]["title"],
                            f"relevance_distance={distance_array_filtered[single_doc_index]}",
                            # single_doc["metadata"]["title"],
                            property_name,
                            f"{firstname} {lastname}'s {property_details['title']}",
                        ],
                        callbacks=[langfuse_handler],
                    )

Marc•13mo ago

if the tags show up in metadata, would this be helpful? (langfuse) metadata

Luke Skywalker•13mo ago

well htey do, but I would love to be able to filter through them in the ui.

Luke Skywalker•13mo ago

Marc•13mo ago

basically tags contains xyz makes sense not P0 right now but might get to it over the next days

Marc•13mo ago

can you raise an issue for this? https://github.com/langfuse/langfuse/issues

GitHub

Issues · langfuse/langfuse

Open source observability and analytics for LLM applications - Issues · langfuse/langfuse

Luke Skywalker•13mo ago

for adding metadata to new spans?

Marc•13mo ago

Oh got it Have you tried the langchain metadata instead of the tags? metadata: Optional[Dict[str, Any]] = None """Optional metadata associated with the chain. Defaults to None. This metadata will be associated with each call to this chain, and passed as arguments to the handlers defined in callbacks. You can use these to eg identify a specific instance of a chain with its use case.

Luke Skywalker•13mo ago

https://github.com/langfuse/langfuse/issues/202

GitHub

feat: Add metadata to manually created spans / observations · Issue...

Describe the feature you'd like to request Would love to be able to set metadata fields on manually created spans: Describe the solution you'd like to see langfusehandler.setNextSpan(next...

Luke Skywalker•13mo ago

ui no, lets see

Marc•13mo ago

do you use langchain metadata already?

Marc•13mo ago

the callback handler should merge tags and metadata into metadata in langfuse https://github.com/langfuse/langfuse-python/blob/752c1937006d3a5cfef93b21617d9164bbc825db/langfuse/callback.py#L545

GitHub

langfuse-python/langfuse/callback.py at 752c1937006d3a5cfef93b21617...

Langfuse Python SDK. Contribute to langfuse/langfuse-python development by creating an account on GitHub.

Luke Skywalker•13mo ago

nope only tags.

extracted_datapoint = extract_datapoint_chain.run(
    query_extract,
    metadata={"foo": "bar"},
    callbacks=[langfuse_handler],
)

extracted_datapoint = extract_datapoint_chain.run(
    query_extract,
    metadata={"foo": "bar"},
    callbacks=[langfuse_handler],
)

Like this?

Marc•13mo ago

and then you can use the filter in the Langfuse UI to select by e.g. extract_model = "extract_data_default"