smolm-autoreg-bpe-counterfactual-babylm-only_random_removal-1e-3

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

• Loss: 3.4056
• 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: 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
3.6071	1.0	18588	3.7805	0.3590
3.3943	2.0	37176	3.5796	0.3806
3.2625	3.0	55764	3.4678	0.3915
3.1838	4.0	74352	3.3962	0.3998
3.1277	5.0	92940	3.3849	0.4017
3.0813	6.0	111528	3.3874	0.4040
3.0519	7.0	130116	3.3394	0.4079
3.0181	8.0	148704	3.3441	0.4085
2.9888	9.0	167292	3.3545	0.4088
2.9602	10.0	185880	3.3501	0.4088
2.942	11.0	204468	3.3509	0.4095
2.9174	12.0	223056	3.3709	0.4093
2.8989	13.0	241644	3.3608	0.4107
2.8757	14.0	260232	3.3651	0.4101
2.8506	15.0	278820	3.3638	0.4109
2.8373	16.0	297408	3.3724	0.4107
2.8195	17.0	315996	3.3819	0.4108
2.7983	18.0	334584	3.3819	0.4110
2.7786	19.0	353172	3.3970	0.4103
2.7635	20.0	371760	3.4056	0.4103

Framework versions

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