Skip to content

fix(asr): resolve native models from the shared catalog - #133

Open
Alphaxiaoteng wants to merge 1 commit into
0xsline:mainfrom
Alphaxiaoteng:fix/large-v3-native-asr-model
Open

fix(asr): resolve native models from the shared catalog#133
Alphaxiaoteng wants to merge 1 commit into
0xsline:mainfrom
Alphaxiaoteng:fix/large-v3-native-asr-model

Conversation

@Alphaxiaoteng

Copy link
Copy Markdown
Contributor

Summary

  • resolve desktop-native GGML filenames from the shared ASR model catalog
  • preserve the existing small and medium filename fallbacks for compatible local caches
  • add executable coverage for every supported tier and unknown model IDs

Problem

The shared catalog already downloads a GGML companion for Whisper Large v3 Turbo, but the desktop worker maintained a separate model map that did not include its model ID. As a result, a complete Large v3 Turbo download could not be selected by the native worker and the app fell back to the browser ONNX path.

This is related to #120 and specifically addresses the Large v3 Turbo desktop-native routing gap. It does not claim to fix the separate Medium ONNX load error or interrupted-download reset reported in that issue.

Verification

  • regression A/B: the old four-entry mapping fails the new Large v3 Turbo assertion; the fix passes all five tiers plus the unknown-model boundary
  • npm test
  • npm run lint
  • npm run build
  • npm run verify:desktop-inference
  • npm run verify:speech-providers
  • npm run verify:agent-skill

No UI behavior changed, so no screenshot is applicable. Developed with Codex assistance and independently reviewed and tested.

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