You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@tooling.apache.org by "KHARSHAVARDHAN-eng (via GitHub)" <gi...@apache.org> on 2026/09/16 03:36:38 UTC
[PR] test: preserve chat_template_kwargs through proxy (#31) (tooling-llmao)
KHARSHAVARDHAN-eng opened a new pull request, #36:
URL: https://github.com/apache/tooling-llmao/pull/36
## Summary
- Add a regression test for #31
- Verify `chat_template_kwargs.enable_thinking` survives the LiteLLM Proxy HTTP path
- Cover both `enable_thinking=true` and `enable_thinking=false`
- Verify the parameter reaches the downstream backend unchanged
## Testing
- `uv run pytest tests/test_chat_template_kwargs.py` — 2 passed
- `git diff --check` — passed
- Full suite: 93 passed, 5 unrelated pre-existing failures
Ref #31
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tooling.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@tooling.apache.org
For additional commands, e-mail: commits-help@tooling.apache.org
Re: [PR] test: preserve chat_template_kwargs through proxy (#31) (tooling-llmao)
Posted by "andrewmusselman (via GitHub)" <gi...@apache.org>.
andrewmusselman merged PR #36:
URL: https://github.com/apache/tooling-llmao/pull/36
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tooling.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@tooling.apache.org
For additional commands, e-mail: commits-help@tooling.apache.org
Re: [PR] test: preserve chat_template_kwargs through proxy (#31) (tooling-llmao)
Posted by "andrewmusselman (via GitHub)" <gi...@apache.org>.
andrewmusselman commented on PR #36:
URL: https://github.com/apache/tooling-llmao/pull/36#issuecomment-5735062563
@KHARSHAVARDHAN-eng
Thanks for picking this up — and the test is worth merging on its own merits,
but it should pass, because #31 turned out not to be a bug.
The parameter was surviving the proxy all along. What was wrong was the check:
LiteLLM returns reasoning as `reasoning_content` on the message, and
duplicates it under `provider_specific_fields.reasoning`. We were reading
`message.reasoning`, which is always empty — so a working `enable_thinking`
looked identical to one that had been dropped.
Verified on the live fleet: gemma4-26b with `enable_thinking: true` returns
744 characters of `reasoning_content` through the gateway. With `false`, zero
and real content.
So I'll close #31 as invalid. This test is still worth having as a regression
guard — it pins the behaviour so a future LiteLLM upgrade that DOES drop the
param gets caught.
One suggestion if you want to extend it: the same class of mistake is easy to
repeat on the response side. A companion assertion that reasoning comes back
under `reasoning_content` (not `reasoning`) would guard the thing that
actually bit us.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscribe@tooling.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@tooling.apache.org
For additional commands, e-mail: commits-help@tooling.apache.org