smolm-autoreg-bpe-counterfactual_babylm_aann_high_variability_adj-1e-3

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

• Loss: 3.4138
• Accuracy: 0.4095

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: 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
3.5997	1.0	18594	3.7835	0.3590
3.3824	2.0	37188	3.5906	0.3796
3.2596	3.0	55782	3.4868	0.3927
3.1824	4.0	74376	3.4542	0.3968
3.1206	5.0	92970	3.3991	0.4011
3.0864	6.0	111564	3.3910	0.4044
3.0439	7.0	130158	3.3760	0.4060
3.0083	8.0	148752	3.3728	0.4063
2.9832	9.0	167346	3.3599	0.4079
2.9563	10.0	185940	3.3528	0.4086
2.9333	11.0	204534	3.3603	0.4092
2.9142	12.0	223128	3.3724	0.4091
2.8926	13.0	241722	3.3841	0.4086
2.8681	14.0	260316	3.3805	0.4093
2.8499	15.0	278910	3.3840	0.4094
2.8344	16.0	297504	3.4004	0.4092
2.8144	17.0	316098	3.3943	0.4096
2.7909	18.0	334692	3.4081	0.4094
2.7754	19.0	353286	3.4054	0.4096
2.7621	20.0	371880	3.4138	0.4095

Framework versions

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