kanishka/smolm-autoreg-bpe-counterfactual_babylm_aann_dtanns-seed_1024-1e-4

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

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

• Loss: 3.4264
• Accuracy: 0.4056

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
4.0514	1.0	18595	4.2435	0.3101
3.5672	2.0	37190	3.7652	0.3622
3.3933	3.0	55785	3.5859	0.3792
3.2939	4.0	74380	3.5397	0.3863
3.2248	5.0	92975	3.4728	0.3919
3.173	6.0	111570	3.4672	0.3950
3.1332	7.0	130165	3.4249	0.3987
3.0958	8.0	148760	3.4232	0.3998
3.0709	9.0	167355	3.4138	0.4012
3.0426	10.0	185950	3.4269	0.4014
3.0138	11.0	204545	3.4023	0.4037
2.995	12.0	223140	3.4037	0.4035
2.9702	13.0	241735	3.3991	0.4043
2.954	14.0	260330	3.4180	0.4042
2.9299	15.0	278925	3.4060	0.4049
2.9106	16.0	297520	3.4084	0.4049
2.8923	17.0	316115	3.4154	0.4055
2.8795	18.0	334710	3.4195	0.4057
2.8628	19.0	353305	3.4225	0.4057
2.8497	20.0	371900	3.4264	0.4056

Framework versions

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