camembert-math-classification

This model is a fine-tuned version of camembert-base on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.0166
• F1 Macro: 0.9987
• F1 Weighted: 0.9987
• F1 Correct: 1.0
• F1 Partiel: 0.9980
• F1 Incorrect: 0.9980
• Accuracy: 0.9987

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: 2e-05
• train_batch_size: 32
• eval_batch_size: 64
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 0.1
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	F1 Macro	F1 Weighted	F1 Correct	F1 Partiel	F1 Incorrect	Accuracy
0.6561	1.0	184	0.8086	0.4912	0.4912	0.8169	0.0204	0.6364	0.5675
0.1997	2.0	368	0.1530	0.9669	0.9669	0.9902	0.9513	0.9592	0.9669
0.0648	3.0	552	0.0360	0.9960	0.9960	1.0	0.9940	0.9941	0.9960
0.0261	4.0	736	0.0227	0.9974	0.9974	0.9980	0.9960	0.9980	0.9974
0.0147	5.0	920	0.0166	0.9987	0.9987	1.0	0.9980	0.9980	0.9987

Framework versions

• Transformers 5.0.0
• Pytorch 2.10.0+cu128
• Datasets 5.0.0
• Tokenizers 0.22.2