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 14, 23:45 UTC
[AMD][CI/Build][Bugfix] Guarding CUDA specific functions by ifndef ROCM (#21766)
Signed-off-by: Gregory Shtrasberg <Gregory.Shtrasberg@amd.com>
Gregory Shtrasberg committed
Jul 28, 2025 at 23:35 UTC
12a223ef9bfebcc61e477047dce049495fe8c8a8
1 file changed
+6
-2
csrc/quantization/compressed_tensors/int8_quant_kernels.cu
+6
-2
@@ -1,7 +1,9 @@
1
#include <ATen/cuda/CUDAContext.h>
2
#include <torch/all.h>
3
4
-#include "../per_token_group_quant_8bit.h"
4
+#ifndef USE_ROCM
5
+ #include "../per_token_group_quant_8bit.h"
6
+#endif
7
8
#include <cmath>
9
@@ -339,10 +341,12 @@ void dynamic_scaled_int8_quant(
341
});
342
}
343
344
+#ifndef USE_ROCM
345
void per_token_group_quant_int8(const torch::Tensor& input,
346
torch::Tensor& output_q,
347
torch::Tensor& output_s, int64_t group_size,
348
double eps, double int8_min, double int8_max) {
349
per_token_group_quant_8bit(input, output_q, output_s, group_size, eps,
350
int8_min, int8_max);
348
-}
\ No newline at end of file
351
+}
352
+#endif