smolm-autoreg-bpe-counterfactual-babylm-only_random_removal-seed_1024-1e-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: 3.4109
• Accuracy: 0.4105

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: 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
3.6055	1.0	18588	3.7771	0.3590
3.3861	2.0	37176	3.5828	0.3803
3.2583	3.0	55764	3.5065	0.3913
3.176	4.0	74352	3.4318	0.3973
3.1227	5.0	92940	3.4132	0.4013
3.0828	6.0	111528	3.3847	0.4036
3.0461	7.0	130116	3.3778	0.4051
3.0138	8.0	148704	3.3612	0.4069
2.9878	9.0	167292	3.3629	0.4078
2.9634	10.0	185880	3.3489	0.4093
2.9347	11.0	204468	3.3616	0.4096
2.9136	12.0	223056	3.3726	0.4097
2.8947	13.0	241644	3.3682	0.4099
2.8789	14.0	260232	3.3817	0.4099
2.8559	15.0	278820	3.3847	0.4099
2.8374	16.0	297408	3.3835	0.4102
2.8117	17.0	315996	3.3940	0.4100
2.7969	18.0	334584	3.4024	0.4102
2.7772	19.0	353172	3.4032	0.4105
2.76	20.0	371760	3.4109	0.4105

Framework versions

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