Remove cuda

Signed-off-by: mgoin <[email protected]>
vllm-project · Nov 6, 2024 · ec074f7 · ec074f7
1 parent bd3a238
commit ec074f7
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/vllm/model_executor/layers/quantization/utils/w8a8_utils.py b/vllm/model_executor/layers/quantization/utils/w8a8_utils.py
@@ -7,7 +7,7 @@
 
 # Input scaling factors are no longer optional in _scaled_mm starting
 # from pytorch 2.5. Allocating a dummy tensor to pass as input_scale
-TORCH_DEVICE_IDENTITY = torch.ones(1).cuda()
+TORCH_DEVICE_IDENTITY = torch.ones(1, dtype=torch.float32)
 
 
 def cutlass_fp8_supported() -> bool: