Instructions to use danqiao-cuhk/OPD-mengrui with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use danqiao-cuhk/OPD-mengrui with PEFT:
Task type is invalid.
- Notebooks
- Google Colab
- Kaggle
OPD Distill Adapters
This repository contains three PEFT/LoRA adapter checkpoints from OPD distillation experiments.
Checkpoints
forward_kl/: forward KL setting. This is the same-date bs80 rerun because the bs64 forward KL run did not save a checkpoint.reverse_kl/: reverse KL on-policy bs64/ga2 setting.entropy_js/: entropy JS on-policy bs64/ga2 setting.
Each subdirectory contains an adapter checkpoint with tokenizer files and PEFT adapter config.
Source Runs
forward_kl:test_opd_distill_verl_forward_kl_onpolicy_bs80_ga2_qwen3_8b_to_4b_8gpu_h200shared_cephonly_20260610_1055reverse_kl:test_opd_distill_verl_reverse_kl_onpolicy_bs64_ga2_qwen3_8b_to_4b_8gpu_h200shared_cephonly_20260610_1905entropy_js:test_opd_distill_verl_entropy_js_onpolicy_bs64_ga2_qwen3_8b_to_4b_8gpu_h200shared_cephonly_20260610_1905
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support