Skip to content

Add device and dtype parameters to StableEmbedding #87

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
merged 1 commit into from
Jan 2, 2023
Merged

Add device and dtype parameters to StableEmbedding #87

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
11 changes: 9 additions & 2 deletions bitsandbytes/nn/modules.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,6 +38,8 @@ def __init__(
scale_grad_by_freq: bool = False,
sparse: bool = False,
_weight: Optional[Tensor] = None,
device=None,
dtype=None,
) -> None:
super(StableEmbedding, self).__init__(
num_embeddings,
Expand All @@ -48,8 +50,10 @@ def __init__(
scale_grad_by_freq,
sparse,
_weight,
device,
dtype,
)
self.norm = torch.nn.LayerNorm(embedding_dim)
self.norm = torch.nn.LayerNorm(embedding_dim, device=device)
GlobalOptimManager.get_instance().register_module_override(
self, "weight", {"optim_bits": 32}
)
Expand Down Expand Up @@ -81,7 +85,10 @@ def forward(self, input: Tensor) -> Tensor:
self.sparse,
)

return self.norm(emb)
# always apply layer norm in full precision
emb = emb.to(torch.get_default_dtype())

return self.norm(emb).to(self.weight.dtype)


class Embedding(torch.nn.Embedding):
Expand Down