[](https://github.com/axolotl-ai-cloud/axolotl)
Model description
WitchLM is cool!
Benchmarks
"leaderboard": {
"inst_level_strict_acc,none": 0.33573141486810554,
"inst_level_strict_acc_stderr,none": "N/A",
"inst_level_loose_acc,none": 0.39568345323741005,
"inst_level_loose_acc_stderr,none": "N/A",
"acc_norm,none": 0.3493319496692178,
"acc_norm_stderr,none": 0.005120138265236575,
"acc,none": 0.24418218085106383,
"acc_stderr,none": 0.003916649280281885,
"exact_match,none": 0.04078549848942598,
"exact_match_stderr,none": 0.005354025092648956,
"prompt_level_strict_acc,none": 0.1977818853974122,
"prompt_level_strict_acc_stderr,none": 0.01714125471908492,
"prompt_level_loose_acc,none": 0.25693160813308685,
"prompt_level_loose_acc_stderr,none": 0.018802962575636854,
"alias": "leaderboard"
},
"leaderboard_bbh": {
"acc_norm,none": 0.3591390383613956,
"acc_norm_stderr,none": 0.0058684522608536275,
"alias": " - leaderboard_bbh"
},
"leaderboard_gpqa": {
"acc_norm,none": 0.29194630872483224,
"acc_norm_stderr,none": 0.013178882651123217,
"alias": " - leaderboard_gpqa"
},
"leaderboard_ifeval": {
"prompt_level_strict_acc,none": 0.1977818853974122,
"prompt_level_strict_acc_stderr,none": 0.01714125471908492,
"inst_level_strict_acc,none": 0.33573141486810554,
"inst_level_strict_acc_stderr,none": "N/A",
"prompt_level_loose_acc,none": 0.25693160813308685,
"prompt_level_loose_acc_stderr,none": 0.018802962575636854,
"inst_level_loose_acc,none": 0.39568345323741005,
"inst_level_loose_acc_stderr,none": "N/A",
"alias": " - leaderboard_ifeval"
},
"leaderboard_math_hard": {
"exact_match,none": 0.04078549848942598,
"exact_match_stderr,none": 0.005354025092648956,
"alias": " - leaderboard_math_hard"
},
"leaderboard_mmlu_pro": {
"acc,none": 0.24418218085106383,
"acc_stderr,none": 0.003916649280281885,
"alias": " - leaderboard_mmlu_pro"
},
"leaderboard_musr": {
"acc_norm,none": 0.36507936507936506,
"acc_norm_stderr,none": 0.01715613678641816,
"alias": " - leaderboard_musr"
}
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 4
- eval_batch_size: 4
- seed: 42
- distributed_type: multi-GPU
- num_devices: 4
- gradient_accumulation_steps: 4
- total_train_batch_size: 64
- total_eval_batch_size: 16
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 50
- num_epochs: 5
Framework versions
- Transformers 4.44.0
- Pytorch 2.3.1+cu121
- Datasets 2.20.0
- Tokenizers 0.19.1