(docs) Update W&B integration with LiteLLM #8052
Conversation
|
The latest updates on your projects. Learn more about Vercel for Git ↗︎
|
|
|
||
| Install W&B Weave | ||
| ```shell | ||
| pip install weave |
There was a problem hiding this comment.
@ayulockin i'm confused - why is the weave sdk required here?
If this is patching litellm then that would be bad, as it can cause unexpected errors.
It would be preferred if weave was written as a customlogger, and did not have an sdk requirement (e.g. using pure httpx like langsmith - https://github.com/BerriAI/litellm/blob/9644e197f760c5cb87ae7662dbedbbce1b04c6a8/litellm/integrations/langsmith.py)
There was a problem hiding this comment.
Hey @krrishdholakia, weave is required because we are patching litellm. It happens when we do weave.init().
It's a light patching and is basically decorating the completion and acompletion functions with weave.op() (docs).
We would love to write a custom logger but for now this patching based integration lives in our weave repo and this PR is meant to update the outdated product feature. We no longer support "Traces" product which the current documentation is showing and cause confusion for users.
We also have a documentation on our site here if you wanna take a look: https://weave-docs.wandb.ai/guides/integrations/litellm and this PR aligns the present capability on both docs.
can you point me to how weave works with litellm? Code would be helpful here
I am sharing the line where the fn is decorated with weave.op() but the whole integration is living in this single file.
|
@ayulockin can you point me to how |
New W&B Weave instrumentation with LiteLLM
The current documentation showing W&B's integration with LiteLLM is outdated. We do not support this feature anymore and opened this PR to update the same.
W&B Weave is our new product for GenAI workflows. We have an auto logging integration in our weave codebase. This PR adds/updates the documentation to reflect the same to avoid any/further confusion.
The PR also adds a
print_verbosestatement indicating that theWeightsBiasesLoggeris deprecated. Do let me know if there is a better way to flag deprecation warning to the users.Type
📖 Documentation