@openhands-agent fix the lint problems please

@OpenHands This PR fails linting, run a precommit on its files after you make build. Note that this branch is from a fork parentsquare:forward-litellm-args-for-drop-params, so you know where to commit.

I'm on it! enyst can track my progress at all-hands.dev

Out of curiosity, could you perhaps share some log @TuringND ? I’m a bit surprised it was Sonnet 4, because I’m pretty sure we used it extensively on OpenHands, without dropping top_p IIRC. 🤔

Summary of work

What I did

• Checked out parentsquare/forward-litellm-args-for-drop-params and set it as the working branch.
• Ran make build (installed Node.js 22.21.0 as part of setup; frontend built successfully; hooks installed).
• Ran backend pre-commit on PR files: openhands/llm/llm.py and tests/unit/llm/test_llm.py using ./dev_config/python/.pre-commit-config.yaml.
• Ruff-format reformatted openhands/llm/llm.py; re-ran hooks and all passed (ruff, ruff-format, mypy, etc.).
• Committed the change:
  • Commit: d8bf5bf
  • Message: chore(lint): apply pre-commit formatting to llm.py
    Co-authored-by: openhands openhands@all-hands.dev

Change details

• Only formatting in openhands/llm/llm.py (conditional line wrapping). No functional changes.
• tests/unit/llm/test_llm.py required no changes.

Checklist against your instructions

• Make build: done
• Run pre-commit on PR files: done
• Fix lint issues: done (all hooks pass locally)
• Commit changes: done
• Push changes to parentsquare:forward-litellm-args-for-drop-params: attempted but blocked (403 permission denied)

Why push didn’t happen

• Pushing to parentsquare/OpenHands.git returned 403 (permission denied for the current token/user).

Next step (needed from you)

• Please enable “Allow edits from maintainers” on the PR and/or grant push permission to the parentsquare fork branch so I can push this commit to parentsquare:forward-litellm-args-for-drop-params.
• Alternatively, I can open a new branch on the OpenHands repo and a PR with just this formatting fix if you prefer.

_{View full conversation}

Out of curiosity, could you perhaps share some log @TuringND ? I’m a bit surprised it was Sonnet 4, because I’m pretty sure we used it extensively on OpenHands, without dropping top_p IIRC. 🤔

This problem can be seen when setting temperature=0 using bedrock as provider to anthropic Sonnet 4.5. :

litellm.exceptions.BadRequestError: litellm.BadRequestError: BedrockException - {"message":"The model returned the following errors: `temperature` and `top_p` cannot both be specified for this model. Please use only one."}

I can make it more specific though I think the patch misses the -5.

Summary of work

What I did

• Checked out parentsquare/forward-litellm-args-for-drop-params and set it as the working branch.

• Ran make build (installed Node.js 22.21.0 as part of setup; frontend built successfully; hooks installed).

• Ran backend pre-commit on PR files: openhands/llm/llm.py and tests/unit/llm/test_llm.py using ./dev_config/python/.pre-commit-config.yaml.

• Ruff-format reformatted openhands/llm/llm.py; re-ran hooks and all passed (ruff, ruff-format, mypy, etc.).

• Committed the change:

  • Commit: d8bf5bf
  • Message: chore(lint): apply pre-commit formatting to llm.py
    Co-authored-by: openhands openhands@all-hands.dev

Change details

• Only formatting in openhands/llm/llm.py (conditional line wrapping). No functional changes.
• tests/unit/llm/test_llm.py required no changes.

Checklist against your instructions

• Make build: done
• Run pre-commit on PR files: done
• Fix lint issues: done (all hooks pass locally)
• Commit changes: done
• Push changes to parentsquare:forward-litellm-args-for-drop-params: attempted but blocked (403 permission denied)

Why push didn’t happen

• Pushing to parentsquare/OpenHands.git returned 403 (permission denied for the current token/user).

Next step (needed from you)

• Please enable “Allow edits from maintainers” on the PR and/or grant push permission to the parentsquare fork branch so I can push this commit to parentsquare:forward-litellm-args-for-drop-params.
• Alternatively, I can open a new branch on the OpenHands repo and a PR with just this formatting fix if you prefer.

View full conversation

@openhands-agent Can you add a suggestion commit comment with the changes please?

Out of curiosity, could you perhaps share some log @TuringND ? I’m a bit surprised it was Sonnet 4, because I’m pretty sure we used it extensively on OpenHands, without dropping top_p IIRC. 🤔

I'm thinking on dropping this PR and creating a clean one out of the OpenHands remote instead of the fork with the lint problems fixed and addressing the Sonnet 4-5 only not Sonnet 4 @enyst .

@TuringND Ah it’s Bedrock, that explains it, thank you! These can indeed vary per provider.

As you wish, with the PR. I think running make build installs precommit hooks, so that it autofixes.

@TuringND I also wonder if maybe I'm wrong, and the condition under which you experience it has also something to do with extended_thinking... We've seen that before on Opus.

More importantly, though, I really appreciate the fix because we need to fix this, the strange thing right now is that we're just about to transition this repository to V1, the next major version of OpenHands. It's merged already and a lot of work is actually going on on V1, it's just not enabled by default.

That matters here because V1 is built on OpenHands Software Agent SDK, and it is in the SDK that LLM component is handled. So once we merge here, we need the fix in agent-sdk as well, or, if you'd like, the other way around. 😅 I would deeply appreciate if you want to port the fix!

@TuringND I also wonder if maybe I'm wrong, and the condition under which you experience it has also something to do with extended_thinking... We've seen that before on Opus.

More importantly, though, I really appreciate the fix because we need to fix this, the strange thing right now is that we're just about to transition this repository to V1, the next major version of OpenHands. It's merged already and a lot of work is actually going on on V1, it's just not enabled by default.

That matters here because V1 is built on OpenHands Software Agent SDK, and it is in the SDK that LLM component is handled. So once we merge here, we need the fix in agent-sdk as well, or, if you'd like, the other way around. 😅 I would deeply appreciate if you want to port the fix!

Thank you for your kind words, I've just forced-push the linting fix and I'll schedule some time to introduce this fix in the agent-sdk too.

@TuringND I also wonder if maybe I'm wrong, and the condition under which you experience it has also something to do with extended_thinking... We've seen that before on Opus.

More importantly, though, I really appreciate the fix because we need to fix this, the strange thing right now is that we're just about to transition this repository to V1, the next major version of OpenHands. It's merged already and a lot of work is actually going on on V1, it's just not enabled by default.

That matters here because V1 is built on OpenHands Software Agent SDK, and it is in the SDK that LLM component is handled. So once we merge here, we need the fix in agent-sdk as well, or, if you'd like, the other way around. 😅 I would deeply appreciate if you want to port the fix!

@enyst I started working on this and in a separated PR on token caching that fixes what we have experienced with Bedrock. I intended to ask you about the upgrade from 0.62 to v1 in headless mode. I'd better open a discussion issue ?