smolm-autoreg-bpe-counterfactual_babylm_aann_dtanns-3e-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: 3.4364
• Accuracy: 0.4091

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: 0.0003
• train_batch_size: 32
• eval_batch_size: 64
• seed: 42
• 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
3.7414	1.0	18595	3.9225	0.3456
3.4399	2.0	37190	3.6349	0.3753
3.2964	3.0	55785	3.5194	0.3889
3.2043	4.0	74380	3.4749	0.3959
3.148	5.0	92975	3.4184	0.3999
3.0956	6.0	111570	3.4167	0.4026
3.0622	7.0	130165	3.3907	0.4053
3.031	8.0	148760	3.3933	0.4054
3.0007	9.0	167355	3.3778	0.4071
2.9738	10.0	185950	3.3897	0.4077
2.9478	11.0	204545	3.3942	0.4075
2.9273	12.0	223140	3.3858	0.4081
2.9087	13.0	241735	3.3945	0.4085
2.8808	14.0	260330	3.4029	0.4087
2.8627	15.0	278925	3.4063	0.4091
2.8411	16.0	297520	3.4120	0.4090
2.8248	17.0	316115	3.4117	0.4091
2.8039	18.0	334710	3.4248	0.4091
2.7872	19.0	353305	3.4300	0.4091
2.7729	20.0	371900	3.4364	0.4091

Framework versions

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