docs: correct three docstring defaults that disagree with the signatures

[avalyset]

Motivation

Three docstrings in the repository state parameter defaults that the function signatures do not have. A reader trusting the docstring would expect different behaviour than the code produces. The goal is to make those three lines agree with the signatures they describe.

Modification

File	Parameter	Docstring said	Signature has
opencompass/openicl/icl_evaluator/code_evaluator.py	retry	Defaults to 3	retry: int = 5
opencompass/models/turbomind_api.py	temperature	Defaults to 0.7	temperature: float = 1.0
opencompass/datasets/SciReasoner/Mol_Instructions/normalized_SW_score.py	matrix_name	default: BLOSUM62	matrix_name='BLOSUM45'

The matrix_name one is the most consequential of the three: BLOSUM62 and BLOSUM45 are different substitution matrices, so a reader trusting the docstring would expect different alignment scores than the function actually produces.

Docstring text only — three lines, no behaviour changes, and no other issues in these files were touched.

Checked and left alone

While looking for this class of drift I also examined four other candidates and found the docstrings correct:

• opencompass/utils/network.py — host reads (default: Google DNS "8.8.8.8"), which matches host: str = '8.8.8.8'.
• opencompass/summarizers/llm_compression.py and opencompass/summarizers/subjective/compass_arena_bradley_terry.py — time_str reads "Defaults to datetime.now().strftime('%Y%m%d_%H%M%S')", wrapped across two lines, matching the signature exactly.
• opencompass/openicl/icl_evaluator/lm_evaluator.py — seed states no default at all, so there is nothing to disagree with. Its signature default is 666, which is undocumented, but that is a gap rather than a contradiction and is left for a separate change.

One related thing I noticed but did not touch, since it is not a default: turbomind_api.py documents end_str in the Args: block of generate, but end_str is not a parameter of that method — it is read from self.end_str. Happy to fix that here or separately if you would like.

BC-breaking (Optional)

No. The change is docstring text only; no signature, default, or runtime behaviour is altered, so nothing downstream can observe it.

Use cases (Optional)

Not a new feature.

Checklist

Before PR:

• Pre-commit or other linting tools are used to fix the potential lint issues.
• Bug fixes are fully covered by unit tests, the case that causes the bug should be added in the unit tests.
• The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
• The documentation has been modified accordingly, like docstring or example tutorials.

After PR:

• If the modification has potential influence on downstream or other related projects, this PR should be tested with those projects.
• CLA has been signed and all committers have signed the CLA in this PR.

Notes on the boxes left unchecked, so you are not left guessing:

• Unit tests. The diff is three lines of docstring text. There is no behaviour to assert on, so no test was added and none is meaningful here.
• Downstream testing. The modification has no potential influence on downstream projects, for the reason given under BC-breaking, so the condition attached to that item does not arise.

On linting: all three files compile and pass ruff check --select E9,F. The three pre-existing extraneous-f-prefix warnings that ruff --isolated reports in these files are present on main as well and are unrelated to this change.

---

On the CLA checklist item: I could not find a CLA document, bot, or status check in this repository, and recent PRs do not appear to have gone through one. All commits here are authored and committed as eirik@ecodeco.no and are GPG-verified. If there is a CLA process I have missed, point me to it and I will sign it.