smolm-autoreg-bpe-counterfactual-babylm-pipps_removal-seed_1024-1e-3

This model was trained from scratch on the kanishka/counterfactual-babylm-pipps_removal dataset. It achieves the following results on the evaluation set:

• Loss: 3.4120
• Accuracy: 0.4103

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: 1024
• 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
3.5992	1.0	18593	3.7873	0.3590
3.3841	2.0	37186	3.5957	0.3798
3.2517	3.0	55779	3.4368	0.3935
3.1729	4.0	74372	3.4158	0.3977
3.1228	5.0	92965	3.4132	0.4017
3.074	6.0	111558	3.3903	0.4035
3.0396	7.0	130151	3.3731	0.4067
3.0136	8.0	148744	3.3697	0.4065
2.9841	9.0	167337	3.3754	0.4067
2.9561	10.0	185930	3.3766	0.4088
2.9356	11.0	204523	3.3834	0.4089
2.9099	12.0	223116	3.3625	0.4105
2.8924	13.0	241709	3.3680	0.4097
2.8738	14.0	260302	3.3766	0.4103
2.8485	15.0	278895	3.3746	0.4108
2.834	16.0	297488	3.3823	0.4107
2.8108	17.0	316081	3.3894	0.4108
2.7936	18.0	334674	3.4001	0.4101
2.7783	19.0	353267	3.4030	0.4107
2.755	20.0	371860	3.4120	0.4103

Framework versions

• Transformers 4.37.2
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.1