smolm-autoreg-bpe-counterfactual_babylm_anans_new-seed_211-1e-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: 3.4227
• Accuracy: 0.4063

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: 211
• 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
4.0574	1.0	18595	4.2623	0.3095
3.5744	2.0	37190	3.7410	0.3630
3.3998	3.0	55785	3.5874	0.3791
3.2911	4.0	74380	3.5155	0.3873
3.2246	5.0	92975	3.4782	0.3919
3.1723	6.0	111570	3.4440	0.3962
3.1287	7.0	130165	3.4271	0.3987
3.0994	8.0	148760	3.3990	0.4007
3.0668	9.0	167355	3.4112	0.4018
3.0398	10.0	185950	3.3915	0.4033
3.0097	11.0	204545	3.4067	0.4037
2.9924	12.0	223140	3.4117	0.4039
2.9702	13.0	241735	3.3926	0.4054
2.9486	14.0	260330	3.4035	0.4053
2.9284	15.0	278925	3.4107	0.4056
2.9143	16.0	297520	3.4057	0.4061
2.8931	17.0	316115	3.4160	0.4058
2.8785	18.0	334710	3.4139	0.4063
2.8611	19.0	353305	3.4191	0.4062
2.8443	20.0	371900	3.4227	0.4063

Framework versions

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