smolm-autoreg-bpe-counterfactual-babylm-new_regex_aanns_removal-seed_42-1e-3

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

• Loss: 3.3818
• Accuracy: 0.4120

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
3.6013	1.0	18593	3.7859	0.3595
3.3808	2.0	37186	3.5963	0.3806
3.2581	3.0	55779	3.4526	0.3931
3.174	4.0	74372	3.4158	0.3984
3.1218	5.0	92965	3.4018	0.4022
3.0764	6.0	111558	3.3647	0.4060
3.0403	7.0	130151	3.3497	0.4073
3.0123	8.0	148744	3.3577	0.4084
2.9806	9.0	167337	3.3481	0.4096
2.9559	10.0	185930	3.3229	0.4107
2.9341	11.0	204523	3.3348	0.4109
2.9141	12.0	223116	3.3268	0.4113
2.8904	13.0	241709	3.3276	0.4122
2.8682	14.0	260302	3.3432	0.4128
2.8518	15.0	278895	3.3533	0.4120
2.8294	16.0	297488	3.3578	0.4120
2.8079	17.0	316081	3.3555	0.4124
2.7936	18.0	334674	3.3698	0.4121
2.7716	19.0	353267	3.3713	0.4124
2.7573	20.0	371860	3.3818	0.4120

Framework versions

• Transformers 4.38.0
• Pytorch 2.3.1+cu121
• Datasets 2.16.1
• Tokenizers 0.15.2