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

Revert "[Performance] Performance improvements in non-blockwise fp8 CUTLASS MoE (#20762) (#21334)

Browse Source 
Signed-off-by: Ming Yang <minos.future@gmail.com>
This commit is contained in:
Ming Yang
2025-07-21 21:49:01 -07:00
committed by GitHub
parent 90f1e55421
commit e7b2042681
6 changed files with 38 additions and 174 deletions
Download Patch File Download Diff File Expand all files Collapse all files

35
benchmarks/kernels/benchmark_grouped_gemm_cutlass.py
View File

@@ -80,11 +80,6 @@ def bench_run(
a, score, topk, renormalize=False
)
ab_strides1 = torch.full((num_experts,), k, device="cuda", dtype=torch.int64)
ab_strides2 = torch.full((num_experts,), n, device="cuda", dtype=torch.int64)
c_strides1 = torch.full((num_experts,), 2 * n, device="cuda", dtype=torch.int64)
c_strides2 = torch.full((num_experts,), k, device="cuda", dtype=torch.int64)
def run_triton_moe(
a: torch.Tensor,
w1: torch.Tensor,
@@ -116,10 +111,6 @@ def bench_run(
w2: torch.Tensor,
w1_scale: torch.Tensor,
w2_scale: torch.Tensor,
ab_strides1: torch.Tensor,
ab_strides2: torch.Tensor,
c_strides1: torch.Tensor,
c_strides2: torch.Tensor,
topk_weights: torch.Tensor,
topk_ids: torch.Tensor,
per_act_token: bool,
@@ -134,10 +125,6 @@ def bench_run(
topk_ids,
w1_scale,
w2_scale,
ab_strides1,
ab_strides2,
c_strides1,
c_strides2,
per_act_token,
a1_scale=None,
)
@@ -149,10 +136,6 @@ def bench_run(
w2_q: torch.Tensor,
w1_scale: torch.Tensor,
w2_scale: torch.Tensor,
ab_strides1: torch.Tensor,
ab_strides2: torch.Tensor,
c_strides1: torch.Tensor,
c_strides2: torch.Tensor,
topk_weights: torch.Tensor,
topk_ids: torch.Tensor,
):
@@ -167,10 +150,6 @@ def bench_run(
topk_ids,
w1_scale,
w2_scale,
ab_strides1,
ab_strides2,
c_strides1,
c_strides2,
per_act_token,
a1_scale=None,
)
@@ -215,10 +194,6 @@ def bench_run(
w2_q,
w1_scale,
w2_scale,
ab_strides1,
ab_strides2,
c_strides1,
c_strides2,
topk_weights,
topk_ids,
)
@@ -256,10 +231,6 @@ def bench_run(
"w1_scale": w1_scale,
"w2_scale": w2_scale,
"per_act_token": per_act_token,
"ab_strides1": ab_strides1,
"ab_strides2": ab_strides2,
"c_strides1": c_strides1,
"c_strides2": c_strides2,
# cuda graph params
"cutlass_graph": cutlass_graph,
"triton_graph": triton_graph,
@@ -318,10 +289,6 @@ def bench_run(
w2_q,
w1_scale,
w2_scale,
ab_strides1,
ab_strides2,
c_strides1,
c_strides2,
topk_weights,
topk_ids,
per_act_token,
@@ -330,7 +297,7 @@ def bench_run(
results.append(
benchmark.Timer(
stmt="run_cutlass_moe(a, a_scale, w1_q, w2_q, w1_scale, w2_scale, ab_strides1, ab_strides2, c_strides1, c_strides2, topk_weights, topk_ids, per_act_token, num_runs)", # noqa: E501
stmt="run_cutlass_moe(a, a_scale, w1_q, w2_q, w1_scale, w2_scale, topk_weights, topk_ids, per_act_token, num_runs)", # noqa: E501
globals=globals,
label=label,
sub_label=sub_label,