Skip to content

Add Confucius4-R2T2 MLX speech model - #111

Merged
IchenDEV merged 2 commits into
mainfrom
t3code/integrate-confucius4-r2t2-gguf-model
Sep 23, 2026
Merged

IchenDEV merged 2 commits into
mainfrom
t3code/integrate-confucius4-r2t2-gguf-model

Conversation

@IchenDEV

@IchenDEV IchenDEV commented Sep 23, 2026 •

Copy link
Copy Markdown
Owner

Outcome

Adds mlx-community/Confucius4-R2T2-8bit as a selectable local speech model in Utter. The model uses the existing Qwen-compatible MLX runtime. A model-specific 0.5-second silence tail preserves final words when a recording stops mid-sentence.

The model download now uses pinned-revision parallel HTTP ranges through the application's existing ModelCatalog.downloadASR path. Each file must match its pinned size and SHA-256 before the generation is published. Other models retain their Hub download path.

Verification

  • Live application catalog download and deletion passed: 2,479,312,565 bytes in 1,704.575 seconds, complete model published, then removed through deleteASR.
  • Range reconstruction, invalid-response, and corruption tests passed.
  • bash scripts/sdlc-checks.sh, bash scripts/ci-basic-checks.sh, and swift test passed (756 XCTest cases, 16 environment skips, no failures; one Swift Testing case passed).
  • bash scripts/build-app.sh --app-only passed release-style app and Metal artifact verification.
  • Pinned MLX runtime previously passed Chinese, English, and abrupt-stop transcription against the real checkpoint. Light and dark model-management renders are in review-evidence/2026-09-23-confucius4-r2t2-mlx/.

Review and release

The user confirmed the model-license decision. The app links the Chinese and English license text and conversion notice; model weights are downloaded, not bundled. The live test invoked the same catalog method as the UI download button. Native button clicks could not be observed because the computer-use native pipe failed to start; this is recorded in the SDLC verification artifact. This PR does not publish a signed release.

SDLC bundle: docs/sdlc/changes/2026-09-23-confucius4-r2t2-mlx/ (medium risk). Reviewer focus: pinned model manifest and hashes, range retry/cancellation, generation publication, model selection, and license links.

@IchenDEV
IchenDEV marked this pull request as ready for review September 23, 2026 14:57
@IchenDEV
IchenDEV merged commit 60e7ed4 into main Sep 23, 2026
3 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.

1 participant