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
[Bug] Enforce contiguous input for `dynamic_scaled_fp8_quant` and `static_scaled_fp8_quant` (#21773)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Wentao Ye committed
Jul 28, 2025 at 15:55 UTC
e0e58f9729e739d857a5ed0d11fc80ea9aa21087
1 file changed
+3
-2
vllm/_custom_ops.py
+3
-2
@@ -1282,10 +1282,11 @@ def scaled_fp8_quant(
1282
output, input.contiguous(), scale, scale_ub)
1283
else:
1284
scale = torch.zeros(1, device=input.device, dtype=torch.float32)
1285
- torch.ops._C.dynamic_scaled_fp8_quant(output, input, scale)
1285
+ torch.ops._C.dynamic_scaled_fp8_quant(output, input.contiguous(),
1286
+ scale)
1287
else:
1288
assert scale.numel() == 1, f"{scale.shape}"
1288
- torch.ops._C.static_scaled_fp8_quant(output, input, scale)
1289
+ torch.ops._C.static_scaled_fp8_quant(output, input.contiguous(), scale)
1290
1291
return output, scale
1292