smolm-autoreg-bpe-counterfactual-babylm-pipps-random_removal-seed_211-1e-3

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

• Loss: 3.3976
• Accuracy: 0.4113

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
3.6046	1.0	18592	3.7833	0.3594
3.3837	2.0	37184	3.5854	0.3805
3.26	3.0	55776	3.4488	0.3931
3.1824	4.0	74368	3.4153	0.3986
3.1238	5.0	92960	3.3853	0.4028
3.0837	6.0	111552	3.3512	0.4060
3.0442	7.0	130144	3.3564	0.4065
3.0168	8.0	148736	3.3438	0.4083
2.9792	9.0	167328	3.3495	0.4090
2.9607	10.0	185920	3.3579	0.4091
2.9363	11.0	204512	3.3420	0.4116
2.9148	12.0	223104	3.3631	0.4106
2.893	13.0	241696	3.3609	0.4106
2.8729	14.0	260288	3.3806	0.4101
2.8543	15.0	278880	3.3685	0.4112
2.8352	16.0	297472	3.3734	0.4119
2.8131	17.0	316064	3.3759	0.4115
2.7949	18.0	334656	3.3842	0.4111
2.7756	19.0	353248	3.3893	0.4115
2.7607	20.0	371840	3.3976	0.4113

Framework versions

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