Skip to content

fix: enabling block-by-block evaluation for granite-3.x-models #165

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
chichun-charlie-liu merged 3 commits into from
Jul 24, 2025
Merged

fix: enabling block-by-block evaluation for granite-3.x-models #165

Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
bf53238
fix: Enabled eval_llm_1GPU for granite-3.x-models
bayo-ibm Jul 22, 2025
6c1da4c
fix: updated ways of getting obtaining logits_scaling
bayo-ibm Jul 23, 2025
3d5b342
fix: added comments to the logits scaling calculation in eval_llm_1GPU
bayo-ibm Jul 24, 2025
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
5 changes: 2 additions & 3 deletions fms_mo/utils/eval_utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -98,6 +98,7 @@ def eval_llm_1GPU(qcfg, model, test_dataset, pre_cache_func=None, **kwargs): #
logger.info("All blocks are computed for evaluation")

nlls = []
logits_scaling = getattr(model.config, "logits_scaling", 1)
# for i, data_mb in enumerate(dloader): #if using dloader.
for i in tqdm(range(qcfg["n_samples"]), desc="Final Evaluating..."):
hidden_states = qcfg["cached_input"][i].to(dev)
Expand All @@ -106,9 +107,7 @@ def eval_llm_1GPU(qcfg, model, test_dataset, pre_cache_func=None, **kwargs): #
hidden_states = ln_f(hidden_states)
lm_head.to(dev)
lm_logits = lm_head(hidden_states)

if model.config.model_type == "granite":
lm_logits /= model.config.logits_scaling
lm_logits /= logits_scaling

# Shift so that tokens < n predict n
shift_logits = lm_logits[:, :-1, :].contiguous().float()
Expand Down
Loading