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

[Performance] Move apply_w8a8_block_fp8_linear to an op class (#24666)

Browse Source 
Signed-off-by: ElizaWszola <ewszola@redhat.com>
Signed-off-by: ElizaWszola <elizaw.9289@gmail.com>
Signed-off-by: Luka Govedič <lgovedic@redhat.com>
Signed-off-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
Co-authored-by: Luka Govedič <ProExpertProg@users.noreply.github.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
Co-authored-by: Luka Govedič <lgovedic@redhat.com>
This commit is contained in:
ElizaWszola
2025-09-23 21:03:10 +02:00
committed by GitHub
parent 8c1c81a3de
commit 63400259d0
14 changed files with 341 additions and 201 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
benchmarks/kernels/deepgemm/benchmark_fp8_block_dense_gemm.py
View File

@@ -10,7 +10,7 @@ from vllm import _custom_ops as ops
from vllm.model_executor.layers.quantization.utils.fp8_utils import (
get_col_major_tma_aligned_tensor,
per_token_group_quant_fp8,
w8a8_block_fp8_matmul,
w8a8_triton_block_scaled_mm,
)
from vllm.triton_utils import triton
from vllm.utils.deep_gemm import calc_diff, fp8_gemm_nt, per_block_cast_to_fp8
@@ -59,7 +59,7 @@ def benchmark_shape(m: int,
# === vLLM Triton Implementation ===
def vllm_triton_gemm():
return w8a8_block_fp8_matmul(A_vllm,
return w8a8_triton_block_scaled_mm(A_vllm,
B_vllm,
A_scale_vllm,
B_scale_vllm,