smolm-autoreg-bpe-counterfactual_babylm_aann_indef_articles_with_pl_nouns_removal_new-1e-4

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

• Loss: 3.4333
• 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: 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
4.0479	1.0	18600	4.2935	0.3070
3.5674	2.0	37200	3.7738	0.3611
3.3916	3.0	55800	3.6022	0.3786
3.2862	4.0	74400	3.5242	0.3885
3.2206	5.0	93000	3.4933	0.3925
3.1712	6.0	111600	3.4670	0.3960
3.1308	7.0	130200	3.4515	0.3982
3.0923	8.0	148800	3.4287	0.4002
3.0627	9.0	167400	3.4128	0.4021
3.0371	10.0	186000	3.4146	0.4029
3.0079	11.0	204600	3.4136	0.4033
2.9826	12.0	223200	3.4180	0.4040
2.9648	13.0	241800	3.3980	0.4056
2.9463	14.0	260400	3.4089	0.4059
2.9268	15.0	279000	3.4190	0.4056
2.9079	16.0	297600	3.4242	0.4058
2.8863	17.0	316200	3.4218	0.4062
2.8721	18.0	334800	3.4296	0.4062
2.8514	19.0	353400	3.4306	0.4064
2.8356	20.0	372000	3.4333	0.4063

Framework versions

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