stage_name string | stage_number int64 | stage_type string | model_repo_id string | base_model string | timestamp string | verl_parameter_config dict |
|---|---|---|---|---|---|---|
rl | 1 | verl_rl_training | TAUR-dev/M-1113_newmodels__qwen7b_ct3arg-rl | Qwen/Qwen2.5-7B-Instruct | 2025-11-13T12:03:18.240473 | {
"actor_rollout_ref.actor.fsdp_config.forward_prefetch": true,
"actor_rollout_ref.actor.optim.lr": 0.000001,
"actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu": 1,
"actor_rollout_ref.actor.ppo_mini_batch_size": 32,
"actor_rollout_ref.actor.strategy": "fsdp2",
"actor_rollout_ref.model.enable_activation_... |
Experiment Tracker: 1113_newmodels__qwen7b_ct3arg
Experiment Description: Experiment: 1113_newmodels__qwen7b_ct3arg
Start Time: 2025-11-13T03:34:52.586872
Tracker Dataset: TAUR-dev/D-ExpTracker__1113_newmodels__qwen7b_ct3arg__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__1113_newmodels__qwen7b_ct3arg__v1', 'experiment_metadata')
# Load complete training datasets
sft_data = load_dataset('TAUR-dev/D-ExpTracker__1113_newmodels__qwen7b_ct3arg__v1', 'training_data__sft')
sft_metadata = load_dataset('TAUR-dev/D-ExpTracker__1113_newmodels__qwen7b_ct3arg__v1', 'training_data__sft_metadata')
# Load complete configurations
sft_hyperparams = load_dataset('TAUR-dev/D-ExpTracker__1113_newmodels__qwen7b_ct3arg__v1', 'hyperparameters__sft')
rl_hyperparams = load_dataset('TAUR-dev/D-ExpTracker__1113_newmodels__qwen7b_ct3arg__v1', 'hyperparameters__rl')
# Load stage-specific logs
sft_logs = load_dataset('TAUR-dev/D-ExpTracker__1113_newmodels__qwen7b_ct3arg__v1', 'logs__sft')
rl_logs = load_dataset('TAUR-dev/D-ExpTracker__1113_newmodels__qwen7b_ct3arg__v1', 'logs__rl')
# Load evaluation results with annotations
sft_eval_results = load_dataset('TAUR-dev/D-ExpTracker__1113_newmodels__qwen7b_ct3arg__v1', 'evals_eval_sft')
rl_eval_results = load_dataset('TAUR-dev/D-ExpTracker__1113_newmodels__qwen7b_ct3arg__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 - 1113_newmodels__qwen7b_ct3arg - {stage_name} - {SFT/RL}
Generated by SkillFactory Experiment Management System All artifacts uploaded immediately as stages complete with perfect data provenance
- Downloads last month
- 7