EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[feature] Add Custom Op grouped_matmul_swiglu_quant (#4431)

Browse Source 
This PR introduces the `EXEC_NPU_CMD` macro, serving as an adapter layer
to simplify the invocation of `aclnn` operators on Ascend NPUs.

**Key Changes:**
* **Adapter Layer:** Added `EXEC_NPU_CMD` macro and related dependencies
to standardize `aclnn` calls.
* **Operator Support:** Integrated `grouped_matmul_swiglu_quant` as a
reference implementation to demonstrate the usage of the new macro.

---


- vLLM version: v0.11.2

---------

Signed-off-by: SlightwindSec <slightwindsec@gmail.com>
This commit is contained in:
Slightwind
2025-11-27 21:56:18 +08:00
committed by GitHub
parent 89a1a65300
commit 9fdabb7b60
10 changed files with 1007 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
csrc/torch_binding_meta.cpp
View File

@@ -35,7 +35,7 @@
namespace vllm_ascend {
namespace meta {
const int64_t INT4_NUMS_IN_INT32 = 8;
std::tuple<at::Tensor, at::Tensor> rotary_embedding_meta(
at::Tensor &positions,
at::Tensor &query,
@@ -114,6 +114,22 @@ std::tuple<at::Tensor &, at::Tensor &, at::Tensor &, at::Tensor &> mla_preproces
return {q_out0, kv_cache_out0, q_out1, kv_cache_out1};
}
std::tuple<at::Tensor, at::Tensor, at::Tensor> grouped_matmul_swiglu_quant(
const at::Tensor &x, const at::Tensor &weight, const at::Tensor &weight_scale, const at::Tensor &x_scale,
const at::Tensor &group_list, const c10::optional<at::Tensor> &bias, const c10::optional<at::Tensor> &offset)
{
int m = x.sizes()[0];
int n = weight.sizes()[2];
bool is_a8w4 = x.dtype() == at::kChar && weight.dtype() == at::kInt;
if (is_a8w4) {
n *= INT4_NUMS_IN_INT32;
}
at::Tensor output = at::empty({m, n/2}, x.options().dtype(c10::ScalarType::Char));
at::Tensor output_scale = at::empty({m}, x.options().dtype(c10::ScalarType::Float));
at::Tensor output_offset = at::empty({}, x.options().dtype(c10::ScalarType::Float));
return {output, output_scale, output_offset};
}
} // namespace meta
} // namespace vllm_ascend
@@ -132,5 +148,7 @@ namespace {
ops.impl("sgmv_expand", &vllm_ascend::meta::sgmv_expand_meta);
// MLA preprocess
ops.impl("mla_preprocess", &vllm_ascend::meta::mla_preprocess);
// grouped_matmul_swiglu_quant meta implementation
ops.impl("grouped_matmul_swiglu_quant", &vllm_ascend::meta::grouped_matmul_swiglu_quant);
}
}