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

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

• Loss: 3.4259
• Accuracy: 0.4097

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
3.6017	1.0	18600	3.7683	0.3593
3.3799	2.0	37200	3.5935	0.3790
3.2546	3.0	55800	3.4823	0.3915
3.1737	4.0	74400	3.4548	0.3978
3.1178	5.0	93000	3.4163	0.4014
3.0736	6.0	111600	3.4017	0.4038
3.0385	7.0	130200	3.3798	0.4057
3.0068	8.0	148800	3.3988	0.4060
2.9774	9.0	167400	3.3728	0.4074
2.9558	10.0	186000	3.3695	0.4087
2.9289	11.0	204600	3.3649	0.4094
2.9058	12.0	223200	3.3604	0.4095
2.8805	13.0	241800	3.3801	0.4098
2.8621	14.0	260400	3.3871	0.4095
2.8423	15.0	279000	3.3872	0.4096
2.8216	16.0	297600	3.3996	0.4097
2.8042	17.0	316200	3.3987	0.4101
2.7834	18.0	334800	3.4020	0.4101
2.7643	19.0	353400	3.4199	0.4097
2.7463	20.0	372000	3.4259	0.4097

Framework versions

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