vllm/attention/layer.py · 2a75c6bc91fc39a0b35b6ead9bee8f05ee3e420a · OpenDAS / vllm_cscc · GitLab

"tests/models/multimodal/generation/test_pixtral.py" did not exist on "d31174a4e1ff7ac1efbdb5d89a24f0e477f95cc8"

Find file Blame History Permalink

add VLLM_USE_OPT_ZEROS to replace triton_ (torch.zeros) · e7f2785f
zhuwenwen authored Nov 13, 2025
```
set default_max_num_batched_tokens = 10240
update qwen3_moe of layernorm
off lightop of moe_fused_gate
```
e7f2785f

layer.py 26.7 KB