Skip to content

fix(metax): avoid vllm._C dependency in activation ops for empty builds #325

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
tengqm wants to merge 1 commit into from
Closed

fix(metax): avoid vllm._C dependency in activation ops for empty builds #325

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
21 changes: 7 additions & 14 deletions vllm_fl/dispatch/backends/vendor/metax/impl/activation.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,17 +15,12 @@
# SPDX-License-Identifier: Apache-2.0
# 2026 - Modified by MetaX Integrated Circuits (Shanghai) Co., Ltd. All Rights Reserved.
import torch
from vllm.model_executor.layers.activation import (
SiluAndMul,
GeluAndMul,
)
import torch.nn.functional as F


def silu_and_mul_maca(obj, x: torch.Tensor) -> torch.Tensor:
"""
SiLU activation followed by element-wise multiplication using CUDA.

Uses vLLM's optimized CUDA kernel when available.
SiLU activation followed by element-wise multiplication.

Args:
obj: The calling obj (for interface consistency)
Expand All @@ -34,15 +29,13 @@ def silu_and_mul_maca(obj, x: torch.Tensor) -> torch.Tensor:
Returns:
Output tensor of shape [..., d]
"""
act_fn = SiluAndMul()
return act_fn.forward_cuda(x)
d = x.shape[-1] // 2
return F.silu(x[..., :d]) * x[..., d:]


def gelu_and_mul_maca(obj, x: torch.Tensor) -> torch.Tensor:
"""
GELU activation followed by element-wise multiplication using CUDA.

Uses vLLM's optimized CUDA kernel when available.
GELU activation followed by element-wise multiplication.

Args:
obj: The calling obj (for interface consistency)
Expand All @@ -51,5 +44,5 @@ def gelu_and_mul_maca(obj, x: torch.Tensor) -> torch.Tensor:
Returns:
Output tensor of shape [..., d]
"""
act_fn = GeluAndMul()
return act_fn.forward_cuda(x)
d = x.shape[-1] // 2
return F.gelu(x[..., :d], approximate="none") * x[..., d:]
Loading