Dataset Viewer
model_name_or_path
string | trust_remote_code
bool | stage
string | do_train
bool | finetuning_type
string | deepspeed
string | dataset
string | template
string | cutoff_len
int64 | max_samples
int64 | overwrite_cache
bool | preprocessing_num_workers
int64 | dataloader_num_workers
int64 | disable_tqdm
bool | output_dir
string | logging_steps
int64 | save_steps
int64 | plot_loss
bool | overwrite_output_dir
bool | per_device_train_batch_size
int64 | gradient_accumulation_steps
int64 | learning_rate
float64 | num_train_epochs
int64 | lr_scheduler_type
string | warmup_ratio
float64 | weight_decay
float64 | adam_beta1
float64 | adam_beta2
float64 | bf16
bool | ddp_timeout
int64 | gradient_checkpointing
bool | save_only_model
bool | enable_masked_ranges
bool | save_strategy
string | save_total_limit
int64 | sf_tracker_dataset_id
string | sf_eval_before_training
bool | sf_wandb_project
string | sf_eval_steps
null | run_name
string |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Qwen/Qwen2.5-1.5B-Instruct
| true |
sft
| true |
full
|
/datastor1/mwadhwa/code/skill-factory/thirdparty/LLaMA-Factory/examples/deepspeed/ds_z2_config.json
|
TAUR_dev__D_SFT_C_voting_setup3_1epch_1e6_all_tasks_only_sft_sft_data__sft_train
|
qwen
| 16,384 | 1,000,000 | true | 1 | 0 | false |
/datastor1/mwadhwa/skill_inject_outputs/sf_experiments/skills_in_rl/voting_setup3_1epch_1e6_all_tasks_only_sft/llamafactory/checkpoints
| 10 | 100,000 | true | true | 1 | 1 | 0.000001 | 1 |
cosine
| 0.05 | 0.0001 | 0.9 | 0.95 | true | 180,000,000 | true | true | false |
steps
| 5 |
TAUR-dev/D-ExpTracker__voting_setup3_1epch_1e6_all_tasks_only_sft__v1
| false |
voting_setup3_1epch_1e6_all_tasks_only_sft_sft
| null |
voting_setup3_1epch_1e6_all_tasks_only_sft_sft
|
Experiment Tracker: voting_setup3_1epch_1e6_all_tasks_only_sft
Experiment Description: Simple test experiment for Skill Factory workflows.
Start Time: 2025-08-19T07:51:25.832299
Tracker Dataset: TAUR-dev/D-ExpTracker__voting_setup3_1epch_1e6_all_tasks_only_sft__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__voting_setup3_1epch_1e6_all_tasks_only_sft__v1', 'experiment_metadata')
# Load complete training datasets
sft_data = load_dataset('TAUR-dev/D-ExpTracker__voting_setup3_1epch_1e6_all_tasks_only_sft__v1', 'training_data__sft')
sft_metadata = load_dataset('TAUR-dev/D-ExpTracker__voting_setup3_1epch_1e6_all_tasks_only_sft__v1', 'training_data__sft_metadata')
# Load complete configurations
sft_hyperparams = load_dataset('TAUR-dev/D-ExpTracker__voting_setup3_1epch_1e6_all_tasks_only_sft__v1', 'hyperparameters__sft')
rl_hyperparams = load_dataset('TAUR-dev/D-ExpTracker__voting_setup3_1epch_1e6_all_tasks_only_sft__v1', 'hyperparameters__rl')
# Load stage-specific logs
sft_logs = load_dataset('TAUR-dev/D-ExpTracker__voting_setup3_1epch_1e6_all_tasks_only_sft__v1', 'logs__sft')
rl_logs = load_dataset('TAUR-dev/D-ExpTracker__voting_setup3_1epch_1e6_all_tasks_only_sft__v1', 'logs__rl')
# Load evaluation results with annotations
sft_eval_results = load_dataset('TAUR-dev/D-ExpTracker__voting_setup3_1epch_1e6_all_tasks_only_sft__v1', 'evals_eval_sft')
rl_eval_results = load_dataset('TAUR-dev/D-ExpTracker__voting_setup3_1epch_1e6_all_tasks_only_sft__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 - voting_setup3_1epch_1e6_all_tasks_only_sft - {stage_name} - {SFT/RL}
Generated by SkillFactory Experiment Management System All artifacts uploaded immediately as stages complete with perfect data provenance
- Downloads last month
- 50