opt-babylm2-clean-spacy-32k_seed-42_3e-4

This model was trained from scratch on the kanishka/babylm2-clean-spacy dataset. It achieves the following results on the evaluation set:

• Loss: 3.0190
• Accuracy: 0.4234

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
3.3944	1.0	15543	3.4212	0.3734
3.1245	2.0	31086	3.2037	0.3940
2.9807	3.0	46629	3.0794	0.4073
2.8872	4.0	62172	3.0205	0.4140
2.8286	5.0	77715	2.9885	0.4180
2.779	6.0	93258	2.9699	0.4206
2.7316	7.0	108801	2.9588	0.4222
2.6909	8.0	124344	2.9554	0.4233
2.6504	9.0	139887	2.9544	0.4238
2.6246	10.0	155430	2.9523	0.4244
2.5988	11.0	170973	2.9568	0.4248
2.5639	12.0	186516	2.9595	0.4248
2.5361	13.0	202059	2.9698	0.4248
2.5098	14.0	217602	2.9747	0.4247
2.4899	15.0	233145	2.9792	0.4247
2.4626	16.0	248688	2.9882	0.4244
2.4399	17.0	264231	2.9961	0.4243
2.4186	18.0	279774	3.0051	0.4239
2.3869	19.0	295317	3.0119	0.4237
2.3686	20.0	310860	3.0190	0.4234

Framework versions

• Transformers 4.45.1
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.0