biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

fp8 kv cache support fix for torch.compile (#22758)

Browse Source 
Signed-off-by: Aleksandr Malyshev <maleksan@amd.com>
Signed-off-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com>
Co-authored-by: Aleksandr Malyshev <maleksan@amd.com>
Co-authored-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com>
Co-authored-by: Gregory Shtrasberg <156009573+gshtras@users.noreply.github.com>
This commit is contained in:
Aleksandr Malyshev
2025-09-16 14:27:11 -07:00
committed by GitHub
parent 02d4b85454
commit 3053a22b33
2 changed files with 4 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/v1/attention/backends/triton_attn.py
View File

@@ -361,7 +361,7 @@ class TritonAttentionImpl(AttentionImpl):
key_cache = key_cache.view(self.fp8_dtype)
value_cache = value_cache.view(self.fp8_dtype)
num_tokens, num_heads, head_size = query.shape
assert layer._q_scale == 1.0, \
assert layer._q_scale_float == 1.0, \
"A non 1.0 q_scale is not currently supported."
if current_platform.is_cuda():
# Skip Q quantization on ROCm and XPU, enable this on cuda