87a19d2d00
来源:
1. Chranos/ixformer (GitHub) → ixformer_sdk/ (230 files, 70K lines)
- inference/functions/vllm.py: vllm_moe_topk_softmax 完整实现 (2033 lines)
- inference/functions/moe.py: MoE ops 完整实现 (1380 lines)
- contrib/vllm_flash_attn/: FA2 Python 接口 (1018 lines)
- contrib/tgi/fused_moe.py: TGI fused MoE (429 lines)
- csrc/include/ixformer/: C++ kernel headers + cmake
2. Deep-Spark/xllm (GitHub) → upstream_ref/xllm_latest/ (+15 files)
- npu_torch/qwen3_5_decoder_layer_impl.cpp/.h
- npu_torch/qwen3_5_gated_delta_net.cpp/.h
- npu_torch/qwen3_next_*.cpp/.h (6 files)
- npu_torch/attention.cpp/.h + fused_moe.cpp/.h + CMakeLists.txt
- models/llm/qwen3_5.h + qwen3_5_mtp.h + qwen3_next.h
- models/vlm/qwen3_5.h
调用链完整性:
ixformer_sdk/inference/functions/vllm.py
→ ops.infer.moe_topk_softmax() (C++ 层)
→ 这就是 base 镜像 libixformer.so 里的实现
upstream_ref/xllm_latest/core/layers/ilu/fused_moe.cpp
→ ixformer::infer::topk_softmax() (直接 C++ 调用)
→ ixformer::infer::group_gemm() → 完整 7-step MoE pipeline
30 lines
633 B
Python
30 lines
633 B
Python
import ixformer.inference.functions as ops
|
|
import torch
|
|
|
|
|
|
def gelu_and_mul():
|
|
pass
|
|
|
|
|
|
def gelu_tanh_and_mul():
|
|
pass
|
|
|
|
|
|
def silu_and_mul(input: torch.Tensor, out: torch.Tensor = None) -> torch.Tensor:
|
|
r"""Fused SiLU and Mul operation.
|
|
|
|
Parameters
|
|
----------
|
|
input: torch.Tensor
|
|
Input tensor, shape (..., 2 * hidden_size).
|
|
|
|
out: Optional[torch.Tensor]
|
|
The the output tensor, if specified, the kernel will update this tensor inplace.
|
|
|
|
Returns
|
|
-------
|
|
output: torch.Tensor
|
|
Output tensor, shape (..., hidden_size).
|
|
"""
|
|
return ops.silu_and_mul(input=input, output=out)
|