xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-21 07:35:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Fix a couple PPLX+CUTLASS MoE bugs (#20825)

Browse Source 
Signed-off-by: ElizaWszola <ewszola@redhat.com>
This commit is contained in:
ElizaWszola 2025-07-13 04:39:14 +02:00 committed by GitHub
parent 42d440c22b
commit 3b3b778d4a
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 37 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/model_executor/layers/fused_moe/pplx_prepare_finalize.py
View File

@ -204,7 +204,7 @@ class PplxPrepareAndFinalize(mk.FusedMoEPrepareAndFinalize):
out_expert_x_scale=expert_x_scale, out_expert_x_scale=expert_x_scale,
dp_x=a1q, dp_x=a1q,
dp_x_scale=a1q_scale, dp_x_scale=a1q_scale,
indices=topk_ids, indices=topk_ids.view(dtype=torch.uint32),
bound_m=bound_m, bound_m=bound_m,
) )
@ -249,7 +249,7 @@ class PplxPrepareAndFinalize(mk.FusedMoEPrepareAndFinalize):
topk_weights = torch.ones_like(topk_weights) topk_weights = torch.ones_like(topk_weights)
self.a2a.combine(out_tokens=output, self.a2a.combine(out_tokens=output,
indices=topk_ids, indices=topk_ids.view(dtype=torch.uint32),
weights=topk_weights, weights=topk_weights,
expert_y=fused_expert_output, expert_y=fused_expert_output,
bound_m=bound_m) bound_m=bound_m)

25
vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors_moe.py
View File

@ -737,10 +737,8 @@ class CompressedTensorsW8A8Fp8MoECutlassMethod(CompressedTensorsMoEMethod):
"For FP8 Fused MoE layer, we require either per tensor or " "For FP8 Fused MoE layer, we require either per tensor or "
"channelwise, dynamic per token quantization.") "channelwise, dynamic per token quantization.")
from vllm.model_executor.layers.fused_moe.cutlass_moe import (
cutlass_moe_fp8)
self.topk_indices_dtype = None self.topk_indices_dtype = None
self.fused_experts = cutlass_moe_fp8 # type: ignore self.fused_experts = None # type: ignore
self.disable_expert_map = False self.disable_expert_map = False
def create_weights(self, layer: torch.nn.Module, num_experts: int, def create_weights(self, layer: torch.nn.Module, num_experts: int,
@ -936,7 +934,11 @@ class CompressedTensorsW8A8Fp8MoECutlassMethod(CompressedTensorsMoEMethod):
per_act_token = a1_scale.numel() != 1 if a1_scale is not None else ( per_act_token = a1_scale.numel() != 1 if a1_scale is not None else (
a2_scale.numel() != 1 if a2_scale is not None else False) a2_scale.numel() != 1 if a2_scale is not None else False)
return self.fused_experts( if self.fused_experts is None:
# If no modular kernel is provided, use cutlass_moe_fp8
from vllm.model_executor.layers.fused_moe.cutlass_moe import (
cutlass_moe_fp8)
return cutlass_moe_fp8(
x, x,
layer.w13_weight, layer.w13_weight,
layer.w2_weight, layer.w2_weight,
@ -951,6 +953,21 @@ class CompressedTensorsW8A8Fp8MoECutlassMethod(CompressedTensorsMoEMethod):
a1_scale=a1_scale, a1_scale=a1_scale,
a2_scale=a2_scale, a2_scale=a2_scale,
) )
else:
return self.fused_experts(
x,
layer.w13_weight,
layer.w2_weight,
topk_weights,
topk_ids,
activation=activation,
global_num_experts=global_num_experts,
expert_map=None if self.disable_expert_map else expert_map,
w1_scale=layer.w13_weight_scale,
w2_scale=layer.w2_weight_scale,
a1_scale=layer.w13_input_scale,
a2_scale=layer.w2_input_scale,
)
class CompressedTensorsW8A8Int8MoEMethod(CompressedTensorsMoEMethod): class CompressedTensorsW8A8Int8MoEMethod(CompressedTensorsMoEMethod):