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 8, 22:45 UTC
[Bugfix][ROCm] Include float8_e4m3fnuz in NCCL Dtype Dispatching (#33713)
Signed-off-by: Micah Williamson <micah.williamson@amd.com>
Micah Williamson committed
Feb 4, 2026 at 07:36 UTC
1d367a738e9098ad4af1f6865747914ccd2c65ca
1 file changed
+1
-1
vllm/distributed/device_communicators/pynccl_wrapper.py
+1
-1
@@ -93,7 +93,7 @@ class ncclDataTypeEnum:
93
return cls.ncclFloat64
94
if dtype == torch.bfloat16:
95
return cls.ncclBfloat16
96
- if dtype == torch.float8_e4m3fn:
96
+ if dtype == current_platform.fp8_dtype():
97
return cls.ncclFloat8e4m3
98
raise ValueError(
99
f"Unsupported dtype {dtype}: should be one of "