Skip to content

feat(eval): add regional MiniMax metadata - #142

Merged
ASuresh0524 merged 2 commits into
StarTrail-org:mainfrom
octo-patch:octo/20260812-parameter-refresh-recvrNfTnWILbi
Aug 30, 2026
Merged

feat(eval): add regional MiniMax metadata#142
ASuresh0524 merged 2 commits into
StarTrail-org:mainfrom
octo-patch:octo/20260812-parameter-refresh-recvrNfTnWILbi

Conversation

@octo-patch

Copy link
Copy Markdown
Contributor

Reason: Refresh MiniMax evaluation configuration with selectable regional endpoints and current model metadata.

Summary

  • add global and China OpenAI- and Anthropic-compatible endpoint pairs
  • select endpoint pairs with MINIMAX_API_REGION while preserving independent overrides
  • expose current context, pricing, modality, and thinking metadata for registered models
  • document regional selection and cover the configuration with regression tests

Checks

  • uv run pytest tests/test_eval_model_config.py -q
  • uv run python -m py_compile eval/lib/model_config.py tests/test_eval_model_config.py

@vercel

vercel Bot commented Aug 12, 2026

Copy link
Copy Markdown

@octo-patch is attempting to deploy a commit to the andylizf's projects Team on Vercel.

A member of the Team first needs to authorize it.

Follow-ups to StarTrail-org#142, on top of it. The regional endpoint selection works; these
are the two edges around it.

- An unsupported MINIMAX_API_REGION was rejected before the endpoint overrides
  were applied, so a deployment that pinned both MINIMAX_API_BASE and
  MINIMAX_ANTHROPIC_API_BASE still had to name a region from a list it was not
  using. The README offers those two as independent overrides for gateways and
  proxies; now they are. A region is consulted, and therefore validated, only
  when at least one base was left unset.

- The metadata splat indexed MINIMAX_MODEL_METADATA directly, so registering a
  model in MINIMAX_MODELS without also adding a capability entry raised
  KeyError for every caller of that model. The two dicts have nothing keeping
  them in sync. The registry is now what decides whether a model resolves, and
  the side table is optional — which costs nothing, since run_bench.py (the
  only consumer) reads api_base, api_key and model, never the metadata.

The metadata values themselves are left exactly as they are.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@ASuresh0524
ASuresh0524 merged commit 88eecc3 into StarTrail-org:main Aug 30, 2026
4 of 5 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants