Skip to content

[Bugfix] Enable FP8 KV cache for FlashInfer and Triton backend on non-sm100 GPUs #24577

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Sep 10, 2025
Merged

[Bugfix] Enable FP8 KV cache for FlashInfer and Triton backend on non-sm100 GPUs #24577

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
c35b6cb
enable fp8 kv flashinfer. fix fa2
gau-nernst Sep 10, 2025
2acebb7
enable triton attn
gau-nernst Sep 10, 2025
0ad66ca
fix pre-commit
gau-nernst Sep 10, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions vllm/platforms/cuda.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -530,6 +530,10 @@ def is_kv_cache_dtype_supported(cls, kv_cache_dtype: str,
supported = flash_attn_supports_fp8()
else:
supported = True
elif attention_backend == "FLASHINFER":
supported = True
elif attention_backend == "TRITON_ATTN_VLLM_V1":
supported = cls.supports_fp8()
return supported

@classmethod
Expand Down
6 changes: 5 additions & 1 deletion vllm/v1/attention/backends/flashinfer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -202,7 +202,11 @@ def __init__(self, kv_cache_spec: AttentionSpec, layer_names: list[str],
else:
assert self.kv_cache_spec.dtype == self.model_config.dtype
self.kv_cache_dtype = self.kv_cache_spec.dtype
self.q_data_type = self.kv_cache_dtype

if supports_trtllm_attention()[0]:
self.q_data_type = self.kv_cache_dtype
else:
self.q_data_type = self.model_config.dtype

self._cascade_wrapper = None # Wrapper for cascade attention

Expand Down