apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-23 00:17:25 +08:00
Code Issues Actions 19 Packages Projects Releases Wiki Activity

init (#6642)

Browse Source 
Co-authored-by: gongweibao <gognweibao@baidu.com>
This commit is contained in:
gongweibao
2026-03-04 21:55:31 +08:00
committed by GitHub
parent 5c8f5184d9
commit ddb06ff83f
306 changed files with 40627 additions and 34418 deletions
Download Patch File Download Diff File Expand all files Collapse all files
custom_ops/gpu_ops/cutlass_kernels/moe_gemm/fused_moe_gemm_kernels_bf16_bf16.cu
+3 -2
View File
@@ -23,7 +23,8 @@ namespace phi {
#ifdef PADDLE_CUDA_BF16
template class MoeGemmRunner<
__nv_bfloat16, cutlass::WintQuantTraits<__nv_bfloat16, cutlass::WintQuantMethod::kNone>>;
__nv_bfloat16,
cutlass::WintQuantTraits<__nv_bfloat16, cutlass::WintQuantMethod::kNone>>;
#endif
} // namespace phi
} // namespace phi
custom_ops/gpu_ops/cutlass_kernels/moe_gemm/fused_moe_gemm_kernels_bf16_int2.cu
+3 -2
View File
@@ -24,7 +24,8 @@ namespace phi {
#ifdef PADDLE_CUDA_BF16
template class MoeGemmRunner<
__nv_bfloat16,
cutlass::WintQuantTraits<__nv_bfloat16, cutlass::WintQuantMethod::kWeightOnlyInt2>>;
cutlass::WintQuantTraits<__nv_bfloat16,
cutlass::WintQuantMethod::kWeightOnlyInt2>>;
#endif
} // namespace phi
} // namespace phi
custom_ops/gpu_ops/cutlass_kernels/moe_gemm/fused_moe_gemm_kernels_bf16_int4.cu
+3 -2
View File
@@ -23,7 +23,8 @@ namespace phi {
#ifdef PADDLE_CUDA_BF16
template class MoeGemmRunner<
__nv_bfloat16,
cutlass::WintQuantTraits<__nv_bfloat16, cutlass::WintQuantMethod::kWeightOnlyInt4>>;
cutlass::WintQuantTraits<__nv_bfloat16,
cutlass::WintQuantMethod::kWeightOnlyInt4>>;
#endif
} // namespace phi
} // namespace phi
custom_ops/gpu_ops/cutlass_kernels/moe_gemm/fused_moe_gemm_kernels_bf16_int8.cu
+3 -2
View File
@@ -24,7 +24,8 @@ namespace phi {
#ifdef PADDLE_CUDA_BF16
template class MoeGemmRunner<
__nv_bfloat16,
cutlass::WintQuantTraits<__nv_bfloat16, cutlass::WintQuantMethod::kWeightOnlyInt8>>;
cutlass::WintQuantTraits<__nv_bfloat16,
cutlass::WintQuantMethod::kWeightOnlyInt8>>;
#endif
} // namespace phi
} // namespace phi
custom_ops/gpu_ops/cutlass_kernels/moe_gemm/fused_moe_gemm_kernels_fp16_fp16.cu
+4 -3
View File
@@ -21,7 +21,8 @@
namespace phi {
template class MoeGemmRunner<half,
cutlass::WintQuantTraits<half, cutlass::WintQuantMethod::kNone>>;
template class MoeGemmRunner<
half,
cutlass::WintQuantTraits<half, cutlass::WintQuantMethod::kNone>>;
} // namespace phi
} // namespace phi
custom_ops/gpu_ops/cutlass_kernels/moe_gemm/fused_moe_gemm_kernels_fp16_int2.cu
+3 -2
View File
@@ -22,6 +22,7 @@
namespace phi {
template class MoeGemmRunner<
half, cutlass::WintQuantTraits<half, cutlass::WintQuantMethod::kWeightOnlyInt2>>;
half,
cutlass::WintQuantTraits<half, cutlass::WintQuantMethod::kWeightOnlyInt2>>;
} // namespace phi
} // namespace phi
custom_ops/gpu_ops/cutlass_kernels/moe_gemm/fused_moe_gemm_kernels_fp16_int4.cu
+3 -2
View File
@@ -22,6 +22,7 @@
namespace phi {
template class MoeGemmRunner<
half, cutlass::WintQuantTraits<half, cutlass::WintQuantMethod::kWeightOnlyInt4>>;
half,
cutlass::WintQuantTraits<half, cutlass::WintQuantMethod::kWeightOnlyInt4>>;
} // namespace phi
} // namespace phi
custom_ops/gpu_ops/cutlass_kernels/moe_gemm/fused_moe_gemm_kernels_fp16_int8.cu
+3 -2
View File
@@ -22,6 +22,7 @@
namespace phi {
template class MoeGemmRunner<
half, cutlass::WintQuantTraits<half, cutlass::WintQuantMethod::kWeightOnlyInt8>>;
half,
cutlass::WintQuantTraits<half, cutlass::WintQuantMethod::kWeightOnlyInt8>>;
} // namespace phi
} // namespace phi