Downloads · 30 days
0
namhokaist/appgen-qwen25-50env-grpo-r9-native-checkpoints-20260711
appgen-qwen25-50env-grpo-r9-native-checkpoints-20260711 is a machine learning model from namhokaist. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for verl.
This private repository is an exact training-recovery bundle for global step 75. It is not a directly loadable Transformers inference model.
Downloads · 30 days
0
Access
Public
Updated Jul 12, 2026
Repo size
123 GB
Likes
0
Public
Click a slice to open those files.
.pt99.5 GB · 81%
From the Hugging Face model README
This private repository is an exact training-recovery bundle for global step 75. It is not a directly loadable Transformers inference model.
The bundle contains all eight FSDP model shards, optimizer/scheduler/RNG state,
dataloader state, UED teacher state, the frozen input data view, run metadata,
and a restart script. Restore this folder under the original persistent run root,
inject HF_TOKEN and WANDB_API_KEY into a dedicated tmux environment, and run
resume_grpo_q25_r9_from_latest.sh. VERL will continue from the latest complete
checkpoint while W&B resumes run hwmajf22.
The public, inference-ready actor is uploaded separately by the step-100 watcher.
The original data_view/ records the exact symlink layout used by the run. A
self-contained dereferenced archive is also included so recovery does not depend
on the original absolute link targets. From the recovery bundle root, restore it
with:
cat data_view_dereferenced_tar_parts/data_view.tar.part-* | tar -xf -
Then place the restored data_view/ and global_step_75/ beneath the original
run root before invoking the resume script.