[New Feature] Support W4Afp8 MoE GroupGemm (#3171)

* init * 增加多线程编译 * fix bug * fix bug * code style * 增加fp16 * 将print替换成assert * 修复stmatrix * 减小单测shape * 减小单测shape
2025-10-05 08:37:06 +08:00 · 2025-08-06 10:34:05 +08:00
parent 841e831575
commit 89397516a8
10 changed files with 1455 additions and 2 deletions
--- a/custom_ops/setup_ops.py
+++ b/custom_ops/setup_ops.py
@@ -498,6 +498,8 @@ elif paddle.is_compiled_with_cuda():
        # Hopper optmized mla
        sources += find_end_files("gpu_ops/mla_attn", ".cu")
        sources += ["gpu_ops/flash_mask_attn/flash_mask_attn.cu"]
+        os.system("python utils/auto_gen_w4afp8_gemm_kernel.py")
+        sources += find_end_files("gpu_ops/w4afp8_gemm", ".cu")

    setup(
        name="fastdeploy_ops",