Skip to content

model: fix SWA per-layer freq_base #18526

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
ngxson wants to merge 1 commit into from
Closed

model: fix SWA per-layer freq_base #18526

Show file tree
Hide file tree
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 src/models/gemma3.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -33,7 +33,7 @@ llm_build_gemma3<iswa>::llm_build_gemma3(const llama_model & model, const llm_gr
float freq_base_l = 0.0f;
float freq_scale_l = 0.0f;

if constexpr (iswa) {
if (hparams.is_swa(il)) {
freq_base_l = model.get_rope_freq_base (cparams, il);
freq_scale_l = model.get_rope_freq_scale(cparams, il);
} else {
Expand Down
1 change: 1 addition & 0 deletions src/models/modern-bert.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,6 +26,7 @@ llm_build_modern_bert<iswa>::llm_build_modern_bert(const llama_model & model, co
for (int il = 0; il < n_layer; ++il) {
float freq_base_l = 0.0f;

// TODO @ngxson : do we expect to apply this globally, or dynamically per layer?
if constexpr (iswa) {

@ngxson ngxson Jan 1, 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.

It's quite strange here because modern-bert uses build_attn_inp_no_cache, so iswa has no effects here. Probably a copy-pasting mistake?

@CISC CISC Jan 1, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

It does, check out #15798

@CISC CISC Jan 1, 2026

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

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

Though the templating was probably unnecessary.

freq_base_l = model.get_rope_freq_base(cparams, il);
} else {
Expand Down
2 changes: 1 addition & 1 deletion src/models/plamo3.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,7 +26,7 @@ llm_build_plamo3<iswa>::llm_build_plamo3(const llama_model & model, const llm_gr

float freq_base_l = 0.0f;
float freq_scale_l = 0.0f;
if constexpr (iswa) {
if (hparams.is_swa(il)) {
freq_base_l = model.get_rope_freq_base (cparams, il);
freq_scale_l = model.get_rope_freq_scale(cparams, il);
} else {
Expand Down
Loading