smolm-autoreg-bpe-counterfactual-babylm-only_random_removal-3e-4

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: 3.4106
• Accuracy: 0.4093

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: 0.0003
• 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
3.738	1.0	18588	3.8740	0.3464
3.4419	2.0	37176	3.6257	0.3761
3.2954	3.0	55764	3.5027	0.3881
3.2097	4.0	74352	3.4473	0.3956
3.1499	5.0	92940	3.3972	0.4010
3.1001	6.0	111528	3.3916	0.4024
3.0682	7.0	130116	3.3846	0.4040
3.0328	8.0	148704	3.3477	0.4062
3.0023	9.0	167292	3.3578	0.4078
2.9723	10.0	185880	3.3477	0.4090
2.9535	11.0	204468	3.3459	0.4087
2.9285	12.0	223056	3.3507	0.4092
2.9092	13.0	241644	3.3771	0.4086
2.886	14.0	260232	3.3730	0.4084
2.8603	15.0	278820	3.3764	0.4093
2.8469	16.0	297408	3.3821	0.4095
2.8298	17.0	315996	3.3892	0.4095
2.8082	18.0	334584	3.3956	0.4097
2.7882	19.0	353172	3.4046	0.4094
2.7746	20.0	371760	3.4106	0.4093

Framework versions

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