xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-13 22:15:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[CPU]Improve cpu fused moe perf (#27244)

Browse Source 
Signed-off-by: Zhang Xiangze <Xiangze.Zhang@arm.com>
This commit is contained in:
xiangze-arm 2025-11-06 19:04:18 +08:00 committed by GitHub
parent 59a50afa08
commit c757a15f0f
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 40 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

52
vllm/model_executor/layers/fused_moe/cpu_fused_moe.py
View File

@ -5,6 +5,7 @@ from collections.abc import Callable
import torch import torch
from torch.nn import functional as F from torch.nn import functional as F
from vllm import _custom_ops as ops
from vllm import envs from vllm import envs
@ -237,7 +238,43 @@ class SGLFusedMOE:
class CPUFusedMOE: class CPUFusedMOE:
def __init__(self, layer: torch.nn.Module) -> None: def __init__(self, layer: torch.nn.Module) -> None:
pass use_onednn_mm = ops._supports_onednn and ops.is_onednn_acl_supported()
num_experts = layer.w13_weight.size(0)
has_w13_bias = hasattr(layer, "w13_bias")
has_w2_bias = hasattr(layer, "w2_bias")
layer.gate_up_linear = []
layer.down_linear = []
for i in range(num_experts):
layer_w13_weight = layer.w13_weight[i]
layer_w13_bias = layer.w13_bias[i] if has_w13_bias else None
layer_w2_weight = layer.w2_weight[i]
layer_w2_bias = layer.w2_bias[i] if has_w2_bias else None
if use_onednn_mm:
gate_up_handle = ops.create_onednn_mm(layer_w13_weight.t(), 32)
layer.gate_up_linear.append(
lambda x, handle=gate_up_handle, bias=layer_w13_bias: ops.onednn_mm(
handle, x, bias
)
)
down_handle = ops.create_onednn_mm(layer_w2_weight.t(), 32)
layer.down_linear.append(
lambda x, handle=down_handle, bias=layer_w2_bias: ops.onednn_mm(
handle, x, bias
)
)
else:
layer.gate_up_linear.append(
lambda x, w=layer_w13_weight, b=layer_w13_bias: F.linear(x, w, b)
)
layer.down_linear.append(
lambda x, w=layer_w2_weight, b=layer_w2_bias: F.linear(x, w, b)
)
if use_onednn_mm: # remove weight
layer.w13_weight = torch.nn.Parameter(torch.empty(0), requires_grad=False)
layer.w2_weight = torch.nn.Parameter(torch.empty(0), requires_grad=False)
def __call__( def __call__(
self, self,
@ -287,8 +324,6 @@ class CPUFusedMOE:
outputs = [] outputs = []
start_idx = 0 start_idx = 0
has_w13_bias = hasattr(layer, "w13_bias")
has_w2_bias = hasattr(layer, "w2_bias")
for i, num_tokens in enumerate(tokens_per_expert): for i, num_tokens in enumerate(tokens_per_expert):
end_idx = start_idx + num_tokens end_idx = start_idx + num_tokens
@ -296,19 +331,12 @@ class CPUFusedMOE:
continue continue
tokens_for_this_expert = sorted_tokens[start_idx:end_idx] tokens_for_this_expert = sorted_tokens[start_idx:end_idx]
layer_w13_weight = layer.w13_weight[i] gate_up = layer.gate_up_linear[i](tokens_for_this_expert)
layer_w13_bias = layer.w13_bias[i] if has_w13_bias else None
layer_w2_weight = layer.w2_weight[i]
layer_w2_bias = layer.w2_bias[i] if has_w2_bias else None
gate_up = F.linear(
tokens_for_this_expert, layer_w13_weight, bias=layer_w13_bias
)
if activation == "swigluoai": if activation == "swigluoai":
gate_up = swigluoai_and_mul(gate_up) gate_up = swigluoai_and_mul(gate_up)
else: else:
gate_up = silu_and_mul(gate_up) gate_up = silu_and_mul(gate_up)
expert_out = F.linear(gate_up, layer_w2_weight, bias=layer_w2_bias) expert_out = layer.down_linear[i](gate_up)
outputs.append(expert_out) outputs.append(expert_out)
start_idx = end_idx start_idx = end_idx