Dataset Viewer
stage_name
stringclasses 1
value | stage_number
int64 1
1
| stage_type
stringclasses 1
value | model_repo_id
stringclasses 1
value | base_model
stringclasses 1
value | timestamp
stringdate 2025-08-14 10:01:12
2025-08-20 19:55:36
| verl_parameter_config
dict |
|---|---|---|---|---|---|---|
rl
| 1 |
verl_rl_training
|
TAUR-dev/M-ppo_only_baseline_all_tasks-rl
|
Qwen/Qwen2.5-1.5B-Instruct
|
2025-08-14T10:01:12.370528
|
{
"actor_rollout_ref.actor.optim.lr": 0.000001,
"actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu": 4,
"actor_rollout_ref.actor.ppo_mini_batch_size": 64,
"actor_rollout_ref.model.enable_activation_offload": true,
"actor_rollout_ref.model.enable_gradient_checkpointing": true,
"actor_rollout_ref.model.path": "/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/baselines/verl/prefetched_models/Qwen__Qwen2_5_1_5B_Instruct",
"actor_rollout_ref.model.trust_remote_code": true,
"actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu": 4,
"actor_rollout_ref.rollout.dtype": "bfloat16",
"actor_rollout_ref.rollout.gpu_memory_utilization": 0.6,
"actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu": 4,
"actor_rollout_ref.rollout.tensor_model_parallel_size": 1,
"algorithm.kl_ctrl.kl_coef": 0.001,
"critic.forward_micro_batch_size_per_gpu": 8,
"critic.model.enable_activation_offload": true,
"critic.model.enable_gradient_checkpointing": true,
"critic.model.path": "/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/baselines/verl/prefetched_models/Qwen__Qwen2_5_1_5B_Instruct",
"critic.model.trust_remote_code": true,
"critic.optim.lr": 0.00001,
"critic.ppo_micro_batch_size_per_gpu": 1,
"custom_reward_function.path": "/scratch/10416/zaynesprague/skill_factory_dir/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py",
"custom_reward_function.reward_kwargs.format_score_weight": 0,
"custom_reward_function.reward_kwargs.reward_max": 10,
"custom_reward_function.reward_kwargs.reward_min": 0,
"custom_reward_function.reward_kwargs.transition_penalty_weight": 0,
"data.max_prompt_length": 512,
"data.max_response_length": 4096,
"data.train_batch_size": 1024,
"data.train_files": "/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/baselines/verl/data/train.parquet",
"data.val_files": "/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/baselines/verl/data/test.parquet",
"trainer.default_local_dir": "/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/baselines/verl/checkpoints",
"trainer.experiment_name": "ppo_only_baseline_all_tasks_rl",
"trainer.logger": "[console,wandb]",
"trainer.n_gpus_per_node": 1,
"trainer.nnodes": 4,
"trainer.project_name": "rl_skills__8_13_25",
"trainer.resume_mode": "disable",
"trainer.save_freq": 10,
"trainer.test_freq": 5,
"trainer.total_epochs": 20,
"trainer.val_before_train": true
}
|
rl
| 1 |
verl_rl_training
|
TAUR-dev/M-ppo_only_baseline_all_tasks-rl
|
Qwen/Qwen2.5-1.5B-Instruct
|
2025-08-20T19:55:37.304836
|
{
"actor_rollout_ref.actor.optim.lr": 0.000001,
"actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu": 4,
"actor_rollout_ref.actor.ppo_mini_batch_size": 64,
"actor_rollout_ref.model.enable_activation_offload": true,
"actor_rollout_ref.model.enable_gradient_checkpointing": true,
"actor_rollout_ref.model.path": "/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/baselines/verl/prefetched_models/Qwen__Qwen2_5_1_5B_Instruct",
"actor_rollout_ref.model.trust_remote_code": true,
"actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu": 4,
"actor_rollout_ref.rollout.dtype": "bfloat16",
"actor_rollout_ref.rollout.gpu_memory_utilization": 0.6,
"actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu": 4,
"actor_rollout_ref.rollout.tensor_model_parallel_size": 1,
"algorithm.kl_ctrl.kl_coef": 0.001,
"critic.forward_micro_batch_size_per_gpu": 8,
"critic.model.enable_activation_offload": true,
"critic.model.enable_gradient_checkpointing": true,
"critic.model.path": "/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/baselines/verl/prefetched_models/Qwen__Qwen2_5_1_5B_Instruct",
"critic.model.trust_remote_code": true,
"critic.optim.lr": 0.00001,
"critic.ppo_micro_batch_size_per_gpu": 1,
"custom_reward_function.path": "/scratch/10416/zaynesprague/skill_factory_dir/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py",
"custom_reward_function.reward_kwargs.format_score_weight": 0,
"custom_reward_function.reward_kwargs.reward_max": 10,
"custom_reward_function.reward_kwargs.reward_min": 0,
"custom_reward_function.reward_kwargs.transition_penalty_weight": 0,
"data.max_prompt_length": 512,
"data.max_response_length": 4096,
"data.train_batch_size": 1024,
"data.train_files": "/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/baselines/verl/data/train.parquet",
"data.val_files": "/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/baselines/verl/data/test.parquet",
"trainer.default_local_dir": "/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/baselines/verl/checkpoints",
"trainer.experiment_name": "ppo_only_baseline_all_tasks_rl",
"trainer.logger": "[console,wandb]",
"trainer.n_gpus_per_node": 1,
"trainer.nnodes": 4,
"trainer.project_name": "rl_skills__8_13_25",
"trainer.resume_mode": "disable",
"trainer.save_freq": 10,
"trainer.test_freq": 5,
"trainer.total_epochs": 30,
"trainer.val_before_train": true
}
|
Experiment Tracker: ppo_only_baseline_all_tasks
Experiment Description: Simple test experiment for Skill Factory workflows.
Start Time: 2025-08-20T08:38:19.534746
Tracker Dataset: TAUR-dev/D-ExpTracker__ppo_only_baseline_all_tasks__v1
Stages Completed
Total stages: 1
Models Created
Dataset Configurations
This tracker dataset contains the following configurations with immediate upload as stages complete:
Training Data (Complete Datasets)
Hyperparameters (Complete Configurations)
Logs (Stage-Specific)
Evaluation Results (Complete with Annotations)
Metadata
- experiment_metadata: Timeline and stage information
Usage
Load specific configurations with:
from datasets import load_dataset
# Load experiment metadata
metadata = load_dataset('TAUR-dev/D-ExpTracker__ppo_only_baseline_all_tasks__v1', 'experiment_metadata')
# Load complete training datasets
sft_data = load_dataset('TAUR-dev/D-ExpTracker__ppo_only_baseline_all_tasks__v1', 'training_data__sft')
sft_metadata = load_dataset('TAUR-dev/D-ExpTracker__ppo_only_baseline_all_tasks__v1', 'training_data__sft_metadata')
# Load complete configurations
sft_hyperparams = load_dataset('TAUR-dev/D-ExpTracker__ppo_only_baseline_all_tasks__v1', 'hyperparameters__sft')
rl_hyperparams = load_dataset('TAUR-dev/D-ExpTracker__ppo_only_baseline_all_tasks__v1', 'hyperparameters__rl')
# Load stage-specific logs
sft_logs = load_dataset('TAUR-dev/D-ExpTracker__ppo_only_baseline_all_tasks__v1', 'logs__sft')
rl_logs = load_dataset('TAUR-dev/D-ExpTracker__ppo_only_baseline_all_tasks__v1', 'logs__rl')
# Load evaluation results with annotations
sft_eval_results = load_dataset('TAUR-dev/D-ExpTracker__ppo_only_baseline_all_tasks__v1', 'evals_eval_sft')
rl_eval_results = load_dataset('TAUR-dev/D-ExpTracker__ppo_only_baseline_all_tasks__v1', 'evals_eval_rl')
Models
Registry
All models from this experiment are automatically registered in the SkillFactory Model Registry with:
- Complete training configuration (hyperparameters, datasets, methods)
- Experiment lineage (links back to this tracker dataset)
- Stage-specific metadata (SFT vs RL training details)
- Structured input data references (training datasets and configurations)
Registry entries follow the naming pattern: Model - ppo_only_baseline_all_tasks - {stage_name} - {SFT/RL}
Generated by SkillFactory Experiment Management System All artifacts uploaded immediately as stages complete with perfect data provenance
- Downloads last month
- 88