smolm-autoreg-bpe-counterfactual_babylm_anans_new-seed_1024-1e-4

This model was trained from scratch on the kanishka/counterfactual_babylm_anans_new dataset. It achieves the following results on the evaluation set:

• Loss: 3.4358
• Accuracy: 0.4073

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: 0.0001
• 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
4.0554	1.0	18595	4.2729	0.3076
3.5589	2.0	37190	3.7451	0.3634
3.3868	3.0	55785	3.5863	0.3813
3.2904	4.0	74380	3.5049	0.3888
3.2198	5.0	92975	3.4777	0.3946
3.1698	6.0	111570	3.4524	0.3970
3.1232	7.0	130165	3.4485	0.3993
3.0906	8.0	148760	3.4315	0.4013
3.0612	9.0	167355	3.4062	0.4034
3.0318	10.0	185950	3.4156	0.4043
3.0102	11.0	204545	3.4183	0.4043
2.9841	12.0	223140	3.4149	0.4052
2.9673	13.0	241735	3.4120	0.4061
2.9473	14.0	260330	3.4053	0.4069
2.9271	15.0	278925	3.4140	0.4066
2.9087	16.0	297520	3.4227	0.4067
2.8915	17.0	316115	3.4177	0.4070
2.8719	18.0	334710	3.4238	0.4071
2.8512	19.0	353305	3.4332	0.4071
2.8466	20.0	371900	3.4358	0.4073

Framework versions

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