Skip to content

Add VibeVoice ASR BitNet artifact safeguards - #732

Open
justinchuby wants to merge 4 commits into
justinchuby-vibevoice-offline-asrfrom
justinchuby-add-vibevoice-asr-bitnet
Open

Add VibeVoice ASR BitNet artifact safeguards#732
justinchuby wants to merge 4 commits into
justinchuby-vibevoice-offline-asrfrom
justinchuby-add-vibevoice-asr-bitnet

Conversation

@justinchuby

Copy link
Copy Markdown
Member

Summary

  • pin microsoft/VibeVoice-ASR-BitNet@66e78021ab8f5f06133d1ab421ba4d348bda97c9 and stream its dense F32 safetensors as the staged ONNX conversion source
  • fail closed for the coupled VibeASR.cpp-native I2_S/I8_S GGUF artifacts during local import and Hub preflight
  • record exact source provenance and document that native BitNet/GGUF execution is not claimed

Validation

  • focused integration, GGUF, streaming, builder, and model-coverage tests
  • full non-integration suite
  • repository lint, formatting, and diff checks
  • pinned five-stage package build and live Hub GGUF preflight
  • exact-head GPT-5.6 Terra review: no significant findings

Evidence and limits

The release GGUFs require VibeASR.cpp packed ternary/I8_S layouts and custom CPU kernels, so generic GGUF/affine ONNX conversion is explicitly unsupported. Real-weight L4/L5 transcription, native ternary/scale parity, native runtime performance, and quantization round-trip fidelity are not claimed.

Stacked on #730.

justinchuby and others added 4 commits September 4, 2026 11:25
Identify the pinned I2_S and I8_S formats before generic GGUF parsing so Mobius fails closed rather than producing an inaccurate ONNX conversion. Record exact artifact fingerprints and surface the restriction in Hub preflight.

Co-authored-by: Copilot App <[email protected]>
Signed-off-by: Justin Chu <[email protected]>
Copilot-Session: d55a8ffc-5b22-499b-bf8b-2a70439dd660
Collecting native tensor type IDs must not require a default argument on injected readers or replace established descriptor-shape validation.

Co-authored-by: Copilot App <[email protected]>
Signed-off-by: Justin Chu <[email protected]>
Copilot-Session: d55a8ffc-5b22-499b-bf8b-2a70439dd660
Pin and classify the official dense F32 safetensors source for staged export while rejecting coupled native I2_S/I8_S GGUF artifacts whose execution semantics cannot be preserved.

Co-authored-by: Copilot App <[email protected]>

Copilot-Session: d55a8ffc-5b22-499b-bf8b-2a70439dd660
Signed-off-by: Justin Chu <[email protected]>
Record the pinned dense F32 conversion source and explicit native I2_S/I8_S GGUF rejection, while preserving source-revision metadata through alternate architecture dispatch.

Co-authored-by: Copilot App <[email protected]>

Copilot-Session: d55a8ffc-5b22-499b-bf8b-2a70439dd660
Signed-off-by: Justin Chu <[email protected]>
@github-actions

github-actions Bot commented Sep 4, 2026

Copy link
Copy Markdown

🏗️ Architecture Diff

Comparing f166a78ccb57a4

Model Sub-model Changes Status

No architecture changes detected.


Legend: ⚪ No change · 🔵 Minor (attrs/inits) · 🟡 Moderate (nodes added/removed) · 🔴 Major (interface changed)

@github-actions

github-actions Bot commented Sep 4, 2026

Copy link
Copy Markdown

Performance Comparison

Comparing f166a78ccb57a4

Model Metric Baseline Current Delta
bert (feature-extraction) model_size_bytes 359 KB 359 KB +0.0%
bert (feature-extraction) num_nodes 68 68 +0.0%
falcon model_size_bytes 364 KB 364 KB +0.0%
falcon num_nodes 66 66 +0.0%
gemma2 model_size_bytes 428 KB 428 KB +0.0%
gemma2 num_nodes 105 105 +0.0%
gpt2 model_size_bytes 324 KB 324 KB +0.0%
gpt2 num_nodes 54 54 +0.0%
llama model_size_bytes 425 KB 425 KB +0.0%
llama num_nodes 60 60 +0.0%
llama (static-cache) model_size_bytes 425 KB 425 KB +0.0%
llama (static-cache) num_nodes 56 56 +0.0%
mamba (ssm-text-generation) model_size_bytes 296 KB 296 KB +0.0%
mamba (ssm-text-generation) num_nodes 94 94 +0.0%
phi3 model_size_bytes 421 KB 421 KB +0.0%
phi3 num_nodes 58 58 +0.0%
phi3 (static-cache) model_size_bytes 421 KB 421 KB +0.0%
phi3 (static-cache) num_nodes 54 54 +0.0%
qwen2 model_size_bytes 425 KB 425 KB +0.0%
qwen2 num_nodes 60 60 +0.0%
qwen2 (static-cache) model_size_bytes 425 KB 425 KB +0.0%
qwen2 (static-cache) num_nodes 56 56 +0.0%
qwen3_5_moe (hybrid-text-generation) model_size_bytes 506 KB 506 KB +0.0%
qwen3_5_moe (hybrid-text-generation) num_nodes 265 265 +0.0%
qwen3_5_text (hybrid-text-generation) model_size_bytes 458 KB 458 KB +0.0%
qwen3_5_text (hybrid-text-generation) num_nodes 127 127 +0.0%
qwen3_5_vl (hybrid-qwen-vl) model_size_bytes 977 KB 977 KB +0.0%
qwen3_5_vl (hybrid-qwen-vl) num_nodes 450 450 +0.0%
t5 (seq2seq) model_size_bytes 836 KB 836 KB +0.0%
t5 (seq2seq) num_nodes 176 176 +0.0%
whisper (speech-to-text) model_size_bytes 1008 KB 1008 KB +0.0%
whisper (speech-to-text) num_nodes 128 128 +0.0%

No performance regressions.

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