Skip to content

Fix Lion to use decoupled weight decay in the Triton 32-bit kernel #2001

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
matthewdouglas merged 2 commits into from
Jul 16, 2026
+11 −1
Merged

Fix Lion to use decoupled weight decay in the Triton 32-bit kernel #2001

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ca274a9
Fix Lion to use decoupled weight decay in the Triton 32-bit kernel
eaglstun Jul 14, 2026
934467d
Merge branch 'main' into fix/lion-triton-decoupled-wd-v2
eaglstun Jul 14, 2026
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
12 changes: 11 additions & 1 deletion bitsandbytes/backends/triton/kernels_optim.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -270,7 +270,11 @@ def _optimizer_update_1state_32bit_triton_kernel(
s1_vals = tl.load(state1_ptr + offsets, mask=mask, other=0.0)

g_vals = gnorm_scale * g_vals
if weight_decay > 0.0:
# Coupled (L2) weight decay: fold wd into the gradient. This is correct for
# MOMENTUM/RMSPROP/ADAGRAD, but NOT for LION (id 4), which uses *decoupled*
# (AdamW-style) weight decay applied to the param directly (see the LION branch
# below and Chen et al. 2023). LION is intentionally excluded here.
if OPTIMIZER_ID != 4 and weight_decay > 0.0:
g_vals = g_vals + p_vals * weight_decay

update_scale = 1.0
Expand All @@ -289,6 +293,12 @@ def _optimizer_update_1state_32bit_triton_kernel(
p_vals = p_vals + update_val

elif OPTIMIZER_ID == 4: # LION
# Lion uses decoupled weight decay: shrink the param directly (p *= 1 - lr*wd)
# rather than folding wd into the gradient. Matches the 8-bit blockwise kernel,
# the default/cpu backends, and the Lion paper (Chen et al. 2023).
if weight_decay > 0.0:
p_vals = p_vals * (1.0 - lr * weight_decay)

momentum_update = s1_vals * beta1 + (1.0 - beta1) * g_vals
update_val = update_scale * lr * tl.where(momentum_update > 0, 1.0, tl.where(momentum_update < 0, -1.0, 0.0))
p_vals = p_vals - update_val
Expand Down
Loading