smolm-autoreg-bpe-counterfactual-babylm-only_random_removal-seed_211-1e-3

This model was trained from scratch on the kanishka/counterfactual-babylm-only_random_removal dataset. It achieves the following results on the evaluation set:

• Loss: 3.3755
• Accuracy: 0.4148

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: 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
3.6032	1.0	18588	3.7897	0.3587
3.3863	2.0	37176	3.5718	0.3804
3.2604	3.0	55764	3.4372	0.3939
3.1796	4.0	74352	3.4083	0.3998
3.1262	5.0	92940	3.3492	0.4057
3.0821	6.0	111528	3.3454	0.4090
3.0487	7.0	130116	3.3345	0.4094
3.0128	8.0	148704	3.3277	0.4115
2.9873	9.0	167292	3.3305	0.4121
2.9539	10.0	185880	3.3189	0.4134
2.9369	11.0	204468	3.3453	0.4131
2.9143	12.0	223056	3.3310	0.4134
2.897	13.0	241644	3.3251	0.4152
2.8756	14.0	260232	3.3490	0.4136
2.8543	15.0	278820	3.3548	0.4146
2.8345	16.0	297408	3.3477	0.4150
2.8109	17.0	315996	3.3563	0.4150
2.7959	18.0	334584	3.3613	0.4152
2.7806	19.0	353172	3.3693	0.4150
2.7578	20.0	371760	3.3755	0.4148

Framework versions

• Transformers 4.37.2
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.1