[Kernel] Fix deprecation function warnings squeezellm quant_cuda_kern…

…el (vllm-project#6901)
neuralmagic · Aug 17, 2024 · 51252d4 · 51252d4
1 parent 493f189
commit 51252d4
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/csrc/quantization/squeezellm/quant_cuda_kernel.cu b/csrc/quantization/squeezellm/quant_cuda_kernel.cu
@@ -197,13 +197,13 @@ void squeezellm_gemm(torch::Tensor vec, torch::Tensor mat, torch::Tensor mul,
   const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
   vllm::squeezellm::NUQ4MatMulKernel<<<blocks, threads, 0, stream>>>(
 #ifndef USE_ROCM
-      (half2*)vec.data<at::Half>(),
+      (half2*)vec.data_ptr<at::Half>(),
 #else
       (__half2*)vec.data_ptr<at::Half>(),
 #endif
       mat.data_ptr<int>(),
 #ifndef USE_ROCM
-      (half2*)mul.data<at::Half>(), (__half*)lookup_table.data<at::Half>(),
+      (half2*)mul.data<at::Half>(), (__half*)lookup_table.data_ptr<at::Half>(),
 #else
       (float2*)mul.data_ptr<float>(),
       (__half*)lookup_table.data_ptr<at::Half>(),