Skip to content

Improve RoPE documentation #4

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
Oct 26, 2023
Merged

Improve RoPE documentation #4

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: 8 additions & 1 deletion llm/llama2/position_embeddings.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,6 +29,9 @@ class RotaryPositionalEmbeddings(nn.Module):
Args:
x (tensor): input tensor to which rope is applied

Returns:
torch.Tensor: output tensor with RoPE applied

"""

def __init__(
Expand Down Expand Up @@ -64,9 +67,13 @@ def forward(self, x: torch.Tensor) -> torch.Tensor:
seq_len = x.size(1)
rope_cache = self.cache[:seq_len]

# cast because the reference does
# reshape input; the last dimension is used for computing the output
# cast to float to match the reference implementation
xshaped = x.float().reshape(*x.shape[:-1], -1, 2)

# reshape the cache for broadcasting
rope_cache = rope_cache.view(1, xshaped.size(1), 1, xshaped.size(3), 2)

x_out2 = torch.stack(
[
xshaped[..., 0] * rope_cache[..., 0]
Expand Down