A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 15, 22:15 UTC
[CI/Build] Fix pre-commit errors from #13571 (#13709)
Signed-off-by: Roger Wang <ywang@roblox.com>
Roger Wang committed
Feb 22, 2025 at 16:50 UTC
82e0d601fc9bc4cdbf307e06d3ed6faf9f4dd72e
2 files changed
+7
-5
csrc/quantization/fp4/nvfp4_scaled_mm_entry.cu
+4
-3
@@ -31,7 +31,8 @@ void cutlass_scaled_fp4_mm(torch::Tensor& D, torch::Tensor const& A,
31
#if defined ENABLE_NVFP4 && ENABLE_NVFP4
32
return cutlass_scaled_fp4_mm_sm100a(D, A, B, A_sf, B_sf, alpha);
33
#endif
34
- TORCH_CHECK_NOT_IMPLEMENTED(false, "No compiled nvfp4 mm kernel, vLLM should "
35
- "be compiled using CUDA 12.8 and target "
36
- "compute capability 100 or above.");
34
+ TORCH_CHECK_NOT_IMPLEMENTED(false,
35
+ "No compiled nvfp4 mm kernel, vLLM should "
36
+ "be compiled using CUDA 12.8 and target "
37
+ "compute capability 100 or above.");
38
}
csrc/quantization/fp4/nvfp4_scaled_mm_kernels.cu
+3
-2
@@ -194,8 +194,9 @@ void runGemm(at::Tensor& D, at::Tensor const& A, at::Tensor const& B,
194
at::Tensor const& A_sf, at::Tensor const& B_sf,
195
at::Tensor const& alpha, int64_t m, int64_t n, int64_t k,
196
cudaStream_t stream) {
197
- TORCH_CHECK(false, "Unsupported CUTLASS version. Set VLLM_CUTLASS_SRC_DIR to "
198
- "a CUTLASS 3.8 source directory to enable support.");
197
+ TORCH_CHECK(false,
198
+ "Unsupported CUTLASS version. Set VLLM_CUTLASS_SRC_DIR to "
199
+ "a CUTLASS 3.8 source directory to enable support.");
200
}
201
#endif // defined(CUTLASS_ARCH_MMA_SM100_SUPPORTED)
202