transformers
Pass packed boundary metadata to Qwen3.5 linear-attention fast kernels from data collator
#45034
Open
Open
Pass packed boundary metadata to Qwen3.5 linear-attention fast kernels from data collator #45034
sdharani91
wants to merge
8 commits into
huggingface:main
from
sdharani91:feature_packing_qwen
Pass packed boundary metadata to Qwen3.5 linear-attention fast kernel…
c1a6df8d
Use collator-provided padding-free kwargs in Qwen3.5 linear attention
33979c76
Fix unit test for Qwen 3.5 fast path
77ce949e
Fix Qwen 3.5 unit test activation
70af17f6
Fix Qwen 3.5 unit test
552c4716
Fix Qwen 3.5 unit test-1
f8ede6ab
Fix Qwen 3.5 unit test-2
701a2351
Update modeling qwen 2.5 file with modular_model_converter
2d25605c
Login to write a write a comment.
Login via GitHub