| { |
| "student": "outputs/pruned/glean-0125inst-math-keep25", |
| "teacher": "allenai/OLMoE-1B-7B-0125-Instruct", |
| "training_mode": "on-policy", |
| "kl_direction": "reverse", |
| "dataset": "allenai/Dolci-Instruct-RL", |
| "dataset_sources": "omega,polaris,orz_math,MathSub,DAPO-Math", |
| "max_difficulty": 4, |
| "trajectories": "outputs/teacher_trajectories/dolci_math_curated.jsonl", |
| "trajectory_dataset": "allenai/Dolci-Instruct-RL", |
| "off_policy_frames": "chat", |
| "off_policy_max_seq_len": 2048, |
| "topk_targets": null, |
| "max_loss_tokens": null, |
| "loss_tokens_per_step": 120000, |
| "teacher_device": "cuda:0", |
| "student_device": "cuda:1", |
| "lr": 3e-05, |
| "optimizer": "adamw8bit", |
| "weight_decay": 0.1, |
| "epochs": 1, |
| "prompts_per_step": 272, |
| "group_size": 4, |
| "rollout_batch": 64, |
| "micro_batch": 4, |
| "max_new_tokens": 512, |
| "max_prompt_len": 1024, |
| "warmup_steps": 10, |
| "max_grad_norm": 1.0, |
| "eval_every": 10, |
| "gsm8k_every": 10, |
| "gsm8k_n": 64, |
| "gsm8k_batch": 16, |
| "gsm8k_frames": "chat", |
| "save_every": 50, |
| "out_dir": "outputs/healed/healing_breadth/glean_math_keep25_seed1224_long", |
| "sweep": 500, |
| "wandb": true, |
| "wandb_project": "glean-heal", |
| "wandb_run_name": "heal-glean-math-keep25-seed1224-long-step500", |
| "wandb_run_id": null, |
| "wandb_resume": null, |
| "wandb_mode": "online", |
| "no_wandb_sync": false, |
| "debug": false, |
| "resume_from": "outputs/healed/healing_breadth/glean_math_keep25_seed1224/step0050", |
| "start_step": 50, |
| "no_grad_checkpointing": false, |
| "seed": 1224, |
| "no_teacher_overlap": false, |
| "sync_checkpoints": false, |
| "rollout_engine": "vllm", |
| "vllm_gpu": "GPU-864c54df-0130-7780-e271-8a5551d1733f", |
| "vllm_port": 8377, |
| "vllm_refresh_every": 1, |
| "vllm_serve_bin": "vllm-plugin/.venv25/bin/python", |
| "vllm_gpu_mem_util": 0.85, |
| "vllm_refresh_mode": "reload", |
| "vllm_live_dir": "/dev/shm/glean_keep25_long_1224", |
| "resolved_kl_direction": "reverse" |
| } |
