Skip to content

feat(llm): offer MiniMax M3.1 Flash first in model presets - #227

Open
octo-patch wants to merge 2 commits into
modelscope:mainfrom
octo-patch:octo/minimax-m31-flash-presets
Open

octo-patch wants to merge 2 commits into
modelscope:mainfrom
octo-patch:octo/minimax-m31-flash-presets

Conversation

@octo-patch

@octo-patch octo-patch commented Oct 20, 2026 •

Copy link
Copy Markdown
Contributor

Reason: The MiniMax model picker still offered legacy presets and did not expose MiniMax-M3.1-Flash for transcript-based clipping.

Summary

  • Offer minimax/MiniMax-M3.1-Flash first among MiniMax choices, followed by minimax/MiniMax-M3.
  • Keep custom model entry enabled and preserve explicitly entered legacy/custom IDs, the app-wide default, and existing transport behavior.
  • Document the choices, regional endpoint configuration, and custom entry behavior.
  • Add real OpenAI SDK tests with an offline HTTP transport, and run these regressions in the existing Python 3.12 / Gradio 4.44.1 CI job.

User impact

Users can select the current MiniMax presets directly for transcript-based clipping. Explicit legacy/custom selections remain usable without silent replacement.

Validation

  • python -m pytest -q tests/test_minimax_http.py tests/test_minimax_api.py tests/test_minimax_launch_integration.py tests/test_github_templates.py tests/test_funasr_requirement.py tests/test_installation_docs.py tests/test_cli_stage_help.py tests/test_release_contract.py -k "not dropdown_accepts" — 72 passed, 1 deselected, 2 subtests passed, using Python 3.12.13, OpenAI SDK 3.35.1, and pytest 9.1.1 in an existing environment accessed read-only.
  • python -m py_compile funclip/launch.py funclip/videoclipper.py funclip/utils/subtitle_utils.py tests/test_minimax_http.py and git diff --check.
  • Checked the current model guide linked from the README.
  • Executed the actual launcher callback through the real SDK serializer using an offline HTTP transport: current presets, explicit legacy/custom IDs, global/CN/custom endpoint handling, credential precedence, prompt roles, and preserved content parts.
  • Pinned CI run 39387242932 passed on this PR head with Python 3.12.15, pytest 8.3.5, Gradio 4.44.1, and OpenAI SDK 3.35.1: dependency consistency check passed, existing runtime/installation/CLI checks 14 passed, and complete MiniMax regressions 50 passed, including the rendered dropdown smoke test. The duplicate local pinned install was canceled while still waiting on the shared install gate.

Screenshots or clips

No screenshot is attached. This changes dropdown entries only; the added Gradio smoke test checks rendering and current/legacy/custom value acceptance. Full ASR, video clipping, and paid inference were not run.

Notes for reviewers

The production change is confined to the MiniMax picker entries. HTTP transport, prompt construction, media paths, and release history are unchanged. The local setup intentionally excludes ML/GPU dependencies. The official current language-model list is https://platform.minimax.io/docs/guides/models-intro.md.

Signed-off-by: octo-patch <266937838+octo-patch@users.noreply.github.com>
Signed-off-by: octo-patch <266937838+octo-patch@users.noreply.github.com>
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.

1 participant