Skip to content

[Spyre-Next] [Bug] Fixed dtype for embedding #883

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
bohnstingl merged 1 commit into from
Apr 2, 2026
Merged

[Spyre-Next] [Bug] Fixed dtype for embedding #883

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
9 changes: 6 additions & 3 deletions vllm_spyre_next/vllm_spyre_next/custom_ops/vocab_parallel_embedding.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -87,7 +87,10 @@ def __init__(self, *args, **kwargs):
logger.debug("Building custom VocabParallelEmbedding for Spyre")

self._target_device = torch.device("spyre")
self._target_dtype = None
# The inputs for spyre need to be torch.int64
self._input_target_dtype = torch.int64
# The weights for spyre need to be torch.float16
self._weight_target_dtype = torch.float16
self.maybe_compiled_forward_spyre = self.maybe_compile(self.forward_spyre)

self._layer_name = register_layer(self, "spyre_vocab_parallel_embedding")
Expand Down Expand Up @@ -154,8 +157,8 @@ def _forward_spyre_impl(self, x: torch.Tensor) -> torch.Tensor:
out_device = x.device

out = self.maybe_compiled_forward_spyre(
convert(x, dtype=self._target_dtype, device=self._target_device),
convert(self.weight.data, dtype=self._target_dtype, device=self._target_device),
convert(x, dtype=self._input_target_dtype, device=self._target_device),
convert(self.weight.data, dtype=self._weight_target_dtype, device=self._target_device),
)

# Transfer back to original device and restore original dtype
Expand Down
Loading