Skip to content

Add VibeVoice ASR Streaming 7B - #497

Merged
0xShug0 merged 5 commits into
mainfrom
feature/vibevoice-asr-streaming-7b
Sep 10, 2026
Merged

Add VibeVoice ASR Streaming 7B#497
0xShug0 merged 5 commits into
mainfrom
feature/vibevoice-asr-streaming-7b

Conversation

@0xShug0

@0xShug0 0xShug0 commented Sep 10, 2026

Copy link
Copy Markdown
Owner

Draft PR for early review and conflict tracking.

This adds the VibeVoice ASR Streaming 7B runtime path and GGUF packaging/spec/docs updates. The branch is opened as draft first so review can happen before spending more time on conflict resolution and final validation.

Current validation includes basic CUDA ASR runs for the streaming 7B GGUFs. Speaker-attributed output is still being checked against the official Python baseline before this is marked ready.

@0xShug0
0xShug0 marked this pull request as ready for review September 10, 2026 00:06
@0xShug0

0xShug0 commented Sep 10, 2026

Copy link
Copy Markdown
Owner Author

Updated with a decoder-cache ownership fix for streaming VRAM.

Validation on a 300s Q8 streaming server request:

Case Peak VRAM Streaming RTF Events Words Output
Before 25053 MiB 0.0638 103 814 baseline
After 17951 MiB 0.0456 103 814 byte-identical transcript

Offline VibeVoice ASR four-speaker output also matches the existing baseline exactly.

@0xShug0
0xShug0 merged commit e4b3c9c into main Sep 10, 2026
6 checks passed
@0xShug0
0xShug0 deleted the feature/vibevoice-asr-streaming-7b branch September 10, 2026 02:59
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