revert the MoE dependence (#3230)

3ee62235 · Yineng Zhang · GitHub · 9829e77e · 9829e77e · 9829e77e
Unverified Commit 3ee62235 authored Jan 31, 2025 by Yineng Zhang Committed by GitHub Jan 31, 2025
14 changed files
--- a/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_hopper_input.cu
+++ b/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_hopper_input.cu
--- a/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels.h
+++ b/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels.h
--- a/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels_bf16_bf16.cu
+++ b/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels_bf16_bf16.cu
--- a/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels_bf16_uint4.cu
+++ b/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels_bf16_uint4.cu
--- a/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels_bf16_uint8.cu
+++ b/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels_bf16_uint8.cu
--- a/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels_fp16_fp16.cu
+++ b/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels_fp16_fp16.cu
--- a/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels_fp16_uint4.cu
+++ b/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels_fp16_uint4.cu
--- a/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels_fp16_uint8.cu
+++ b/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels_fp16_uint8.cu
--- a/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels_fp32_fp32.cu
+++ b/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels_fp32_fp32.cu
--- a/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels_fp8_fp8.cu
+++ b/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels_fp8_fp8.cu
--- a/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels_template.h
+++ b/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels_template.h
--- a/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels_template_sm90.h
+++ b/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_gemm_kernels_template_sm90.h
--- a/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_sm90_traits.h
+++ b/sgl-kernel/3rdparty/tensorrt_llm/kernels/cutlass_kernels/moe_gemm/moe_sm90_traits.h
--- a/sgl-kernel/setup.py
+++ b/sgl-kernel/setup.py