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

[Refactor] Remove Duplicate per_block_cast_to_fp8, Remove Dependencies of DeepGEMM (#21787)

Browse Source 
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye
2025-07-31 21:13:27 -04:00
committed by GitHub
parent 0bd409cf01
commit 3700642013
8 changed files with 55 additions and 132 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
tests/kernels/moe/test_deepgemm.py
View File

@@ -69,8 +69,12 @@ def make_block_quant_fp8_weights(
dtype=torch.float32)
for i in range(e):
w1[i], w1_s[i] = per_block_cast_to_fp8(w1_bf16[i])
w2[i], w2_s[i] = per_block_cast_to_fp8(w2_bf16[i])
w1[i], w1_s[i] = per_block_cast_to_fp8(w1_bf16[i],
block_size=block_size,
use_ue8m0=True)
w2[i], w2_s[i] = per_block_cast_to_fp8(w2_bf16[i],
block_size=block_size,
use_ue8m0=True)
return w1, w2, w1_s, w2_s