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

[W8A8 Block Linear Refactor][2/N] Remove W8A8Fp8BlockLinearOp and adopt Fp8 block linear kernel selections. (#33892)

Browse Source 
Signed-off-by: maral <maralbahari.98@gmail.com>
Signed-off-by: Maral <maralbahari.98@gmail.com>
This commit is contained in:
Maral
2026-04-09 08:50:39 +08:00
committed by GitHub
parent 8332078cfd
commit 2e9034c998
35 changed files with 1710 additions and 904 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
tests/compile/passes/test_fusion_attn.py
View File

@@ -66,6 +66,7 @@ class AttentionQuantPatternModel(torch.nn.Module):
self.kv_cache_dtype = kv_cache_dtype
self.device = device
self.vllm_config = vllm_config
self.dtype = vllm_config.model_config.dtype
self.attn = Attention(
num_heads=self.num_qo_heads,
@@ -155,6 +156,7 @@ class TestAttentionFp8StaticQuantPatternModel(AttentionQuantPatternModel):
activation_quant_key=self.quant_key,
weight_quant_key=self.quant_key,
device=self.device,
input_dtype=self.dtype,
)
w = kwargs.get("w")