smolm-autoreg-bpe-counterfactual-babylm-aann-prototypical_only-1e-3

This model was trained from scratch on the kanishka/counterfactual_babylm_prototypical_only dataset. It achieves the following results on the evaluation set:

• Loss: 3.3998
• Accuracy: 0.4126

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.001
• train_batch_size: 32
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 32000
• num_epochs: 20.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
3.6032	1.0	18593	3.7810	0.3584
3.3845	2.0	37186	3.5963	0.3798
3.2554	3.0	55779	3.4491	0.3922
3.1823	4.0	74372	3.4299	0.3975
3.126	5.0	92965	3.3983	0.4004
3.0782	6.0	111558	3.4061	0.4035
3.0476	7.0	130151	3.3689	0.4055
3.0126	8.0	148744	3.3660	0.4087
2.9875	9.0	167337	3.3452	0.4100
2.9573	10.0	185930	3.3508	0.4099
2.9399	11.0	204523	3.3627	0.4095
2.9164	12.0	223116	3.3622	0.4103
2.8939	13.0	241709	3.3502	0.4119
2.872	14.0	260302	3.3635	0.4119
2.8506	15.0	278895	3.3691	0.4118
2.8368	16.0	297488	3.3674	0.4131
2.8123	17.0	316081	3.3751	0.4130
2.7955	18.0	334674	3.3827	0.4134
2.7767	19.0	353267	3.3969	0.4124
2.7592	20.0	371860	3.3998	0.4126

Framework versions

• Transformers 4.36.0
• Pytorch 2.1.1+cu121
• Datasets 2.15.0
• Tokenizers 0.15.0