Commit History
3 days ago
-
Audio encoder: accept the keywords transformers 5.22 passes
Verified signaturetransformers 5.22 calls causal_conv1d_fn with the layer's keywords (sliding_window); like transformers' own torch path, the convolution now ignores them. Without it, an audio clip raised a TypeError on 5.22.431f1d6 -
fuse_projections: view a group already in one storage instead of copying it (#4)
- fuse_projections: view a group already in one storage instead of copying it (87fa2b1240ff228683330003d60f84b7b8d36ec3) Co-authored-by: Chao-Chun (Joe) Hsu <joe32140@users.noreply.huggingface.co>d38ae36 -
Add evaluation results (#3)
- Add evaluation results (9daa78a3882f84a4d716eee0fed50a9fc6c54e51) Co-authored-by: Nathan Habib <SaylorTwift@users.noreply.huggingface.co>80d1b1c -
Accept JSON instructions and option values
Verified signatureQuestions whose instructions or option descriptions are JSON objects, as some Decision Index benchmarks write them, are read as they are; only a media dict is a spoken question.28dd9f7
4 days ago
-
Declare base_model: LiquidAI/LFM3-Omni-2B (#2)
- Declare base_model: LiquidAI/LFM3-Omni-2B (84135af04ae46ca76674bc8a7a89e7fbc4f780fc) Co-authored-by: Richard Suarez <richie-liquid@users.noreply.huggingface.co>c8d48c0
5 days ago
-
d1-Omni-2B
Co-authored-by: mlabonne <mlabonne@users.noreply.huggingface.co> Co-authored-by: Aurelien-Lac <Aurelien-Lac@users.noreply.huggingface.co>4cccbfd
