Skip to content

Fix key rerotation device mismatch 🤖🤖🤖 #242

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

Closed
SimJeg wants to merge 1 commit into from
Closed

Fix key rerotation device mismatch 🤖🤖🤖 #242

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
7 changes: 5 additions & 2 deletions kvpress/presses/key_rerotation_press.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
# SPDX-FileCopyrightText: Copyright (c) 1993-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-FileCopyrightText: Copyright (c) 1993-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.

@maxjeblick maxjeblick Jun 16, 2026

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

Can we make a bulk rewrite to update the whole repo?

@SimJeg SimJeg Jun 16, 2026

Copy link
Copy Markdown
Collaborator Author

Choose a reason for hiding this comment

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

we can do that before the next release

# SPDX-License-Identifier: Apache-2.0


Expand Down Expand Up @@ -70,8 +70,10 @@ def _rerotate_cos_sin(x, inv_freq, selected_positions):
Cosine and sine embeddings, each of shape
``(bsz, num_key_value_heads, n_kept, d)``, matching ``dtype``/``device`` of ``x``.
"""
device = x.device
selected_positions = selected_positions.to(device=device)
inv_freq = inv_freq.to(device=device)
bsz, num_key_value_heads, n_kept = selected_positions.shape
device = selected_positions.device
device_type = x.device.type
dtype = x.dtype
# Original positional indices
Expand Down Expand Up @@ -119,6 +121,7 @@ def rerotate_keys(
The rerotated keys tensor of shape
``(bsz, num_heads, n_kept, d)``.
"""
indices = indices.to(device=keys.device)
new_cos, new_sin = KeyRerotationPress._rerotate_cos_sin(keys, module.rotary_emb.inv_freq, indices)
indices = indices.unsqueeze(-1).expand(-1, -1, -1, module.head_dim)
keys = keys.gather(2, indices).contiguous()
Expand Down
Loading