Skip to content

[Bugfix] Fix Dtypes for Pynccl Wrapper #33030

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
robertgshaw2-redhat merged 5 commits into from
Jan 26, 2026
+8 −2
Merged

[Bugfix] Fix Dtypes for Pynccl Wrapper #33030

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
f0b38ee
updated
Jan 25, 2026
f2afc0f
updated
Jan 25, 2026
5ce9572
int8 -> uint8
Jan 25, 2026
56d0322
Merge branch 'main' into fix-fp8-dtype-sending
robertgshaw2-redhat Jan 26, 2026
cf5aacb
Merge branch 'main' into fix-fp8-dtype-sending
robertgshaw2-redhat Jan 26, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 6 additions & 2 deletions vllm/distributed/device_communicators/pynccl_wrapper.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -72,7 +72,8 @@ class ncclDataTypeEnum:
ncclFloat64 = 8
ncclDouble = 8
ncclBfloat16 = 9
ncclNumTypes = 10
ncclFloat8e4m3 = 10
ncclNumTypes = 11

@classmethod
def from_torch(cls, dtype: torch.dtype) -> int:
Expand All @@ -92,9 +93,12 @@ def from_torch(cls, dtype: torch.dtype) -> int:
return cls.ncclFloat64
if dtype == torch.bfloat16:
return cls.ncclBfloat16
if dtype == torch.float8_e4m3fn:
return cls.ncclFloat8e4m3
raise ValueError(
f"Unsupported dtype {dtype}: should be one of "
f"int8, uint8, int32, int64, float16, float32, float64, bfloat16."
f"int8, uint8, int32, int64, float16, float32, float64, bfloat16,"
" float8e4m3."
)
Comment on lines +96 to 102
Copy link
Copy Markdown
Contributor

@gemini-code-assist gemini-code-assist bot Jan 25, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

high

The codebase, for example in vllm/model_executor/layers/quantization/utils/fp8_utils.py, seems to use both torch.float8_e4m3fn and torch.float8_e4m3fnuz. This function should handle both types to avoid ValueError during collective communication operations with torch.float8_e4m3fnuz tensors. The error message is also updated for clarity.

For more complete FP8 support, you might also consider adding torch.float8_e5m2. This would involve adding ncclFloat8e5m2 to ncclDataTypeEnum and handling torch.float8_e5m2 in this method.

Suggested change
if dtype == torch.float8_e4m3fn:
return cls.ncclFloat8e4m3
raise ValueError(
f"Unsupported dtype {dtype}: should be one of "
f"int8, uint8, int32, int64, float16, float32, float64, bfloat16."
f"int8, uint8, int32, int64, float16, float32, float64, bfloat16,"
" float8e4m3."
)
if dtype in (torch.float8_e4m3fn, torch.float8_e4m3fnuz):
return cls.ncclFloat8e4m3
raise ValueError(
f"Unsupported dtype {dtype}: should be one of "
"int8, uint8, int32, int64, float16, float32, float64, bfloat16, "
"float8_e4m3fn, "
"float8_e4m3fnuz."
)



Expand Down
2 changes: 2 additions & 0 deletions vllm/model_executor/layers/fused_moe/flashinfer_cutlass_prepare_finalize.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -224,6 +224,8 @@ def prepare(
a1q_scale = None

if is_nvfp4 and a1q_scale is not None:
if a1q_scale.element_size() == 1:
a1q_scale = a1q_scale.view(torch.uint8)
a1q_scale = nvfp4_block_scale_interleave(a1q_scale)

return a1q, a1q_scale, None, topk_ids, topk_weights
Expand Down