Skip to content

fix: qwen3_8b_hellaswag_pp_peft recipe #1335

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
ZhiyuLi-Nvidia merged 5 commits into from
Mar 6, 2026
Merged

fix: qwen3_8b_hellaswag_pp_peft recipe #1335

Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
dfc655d
update recipe
ZhiyuLi-Nvidia Feb 19, 2026
e4d250a
Merge branch 'main' into zhiyul/fix_qwen3_8b_hellaswag_pp_peft
ZhiyuLi-Nvidia Mar 3, 2026
33deeed
update
ZhiyuLi-Nvidia Mar 4, 2026
99b56e1
Merge branch 'main' into zhiyul/fix_qwen3_8b_hellaswag_pp_peft
ZhiyuLi-Nvidia Mar 6, 2026
453a48f
Merge branch 'main' into zhiyul/fix_qwen3_8b_hellaswag_pp_peft
ZhiyuLi-Nvidia Mar 6, 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
6 changes: 4 additions & 2 deletions examples/llm_finetune/qwen/qwen3_8b_hellaswag_pp_peft.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -25,6 +25,7 @@ step_scheduler:
ckpt_every_steps: 200
val_every_steps: 100 # will run every x number of gradient steps
num_epochs: 2
max_steps: 100

dist_env:
backend: nccl
Expand All @@ -38,7 +39,6 @@ rng:
model:
_target_: nemo_automodel.NeMoAutoModelForCausalLM.from_pretrained
pretrained_model_name_or_path: Qwen/Qwen3-8B
cache_dir: /nemo-rl/hf_cache/

peft:
_target_: nemo_automodel.components._peft.lora.PeftConfig
Expand Down Expand Up @@ -93,7 +93,9 @@ validation_dataset:

validation_dataloader:
_target_: torchdata.stateful_dataloader.StatefulDataLoader
collate_fn: nemo_automodel.components.datasets.utils.default_collater
collate_fn:
_target_: nemo_automodel.components.datasets.utils.default_collater
pad_seq_len_divisible: 168 # must match the training batch seq_len to avoid static PP shape mismatch

optimizer:
_target_: torch.optim.Adam
Expand Down
Loading