Skip to content

[Bugfix] Correct method call for _set_cos_sin_cache #774

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
May 9, 2025
Merged

[Bugfix] Correct method call for _set_cos_sin_cache #774

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
2 changes: 1 addition & 1 deletion vllm_ascend/ops/rotary_embedding.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -82,7 +82,7 @@ def native_rope_deepseek_forward(self,
offsets: Optional[torch.Tensor] = None,
max_seq_len: Optional[int] = None):
if max_seq_len is not None and max_seq_len > self.max_seq_len:
self._set_cos_sin_cache(max_seq_len, query.device, query.dtype)
Copy link
Collaborator

@ganyi1996ppo ganyi1996ppo May 8, 2025

Choose a reason for hiding this comment

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

Just curious, why this invoke may cause crash issue

Copy link
Collaborator Author

@jianzs jianzs May 8, 2025 •
edited
Loading

Choose a reason for hiding this comment

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

Just curious, why this invoke may cause crash issue

DeepseekScalingRotaryEmbedding doesn't have a _set_cos_sin_cache method. Instead, it uses the self as the first parameter directly when calling the function.

Copy link
Collaborator

@ganyi1996ppo ganyi1996ppo May 9, 2025

Choose a reason for hiding this comment

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

oh, right, we did not patch this function inside the code, thanks for this report and fix

_set_cos_sin_cache(self, max_seq_len, query.device, query.dtype)
if len(key.shape) == 2:
key = key[:, None, :]
# Note: we implement the non neox_style method with shuffle the last dim and neox style
Expand Down