Teloxico/modernbert-4-disfluency-v2

my-modernbert-disfluency

This model is a fine-tuned version of arielcerdap/modernbert-base-multiclass-disfluency on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0586
• Model Preparation Time: 0.0032
• Accuracy: 0.9812
• F1: 0.9513
• Precision: 0.9683
• Recall: 0.9349
• F1 Macro: 0.9698
• Precision Macro: 0.9762
• Recall Macro: 0.9637

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: 1e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 3
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time	Accuracy	F1	Precision	Recall	F1 Macro	Precision Macro	Recall Macro
0.1078	0.3639	500	0.0864	0.0032	0.9723	0.9265	0.9680	0.8884	0.9547	0.9706	0.9406
0.07	0.7278	1000	0.0692	0.0032	0.9781	0.9428	0.9698	0.9173	0.9646	0.9749	0.9551
0.0583	1.0917	1500	0.0651	0.0032	0.9798	0.9474	0.9715	0.9245	0.9675	0.9766	0.9589
0.0471	1.4556	2000	0.0582	0.0032	0.9805	0.9497	0.9656	0.9342	0.9688	0.9748	0.9630
0.0508	1.8195	2500	0.0600	0.0032	0.9803	0.9484	0.9776	0.9209	0.9681	0.9793	0.9578
0.038	2.1834	3000	0.0586	0.0032	0.9812	0.9513	0.9683	0.9349	0.9698	0.9762	0.9637
0.0351	2.5473	3500	0.0614	0.0032	0.9809	0.9503	0.9709	0.9305	0.9692	0.9770	0.9618
0.034	2.9112	4000	0.0616	0.0032	0.9810	0.9505	0.9713	0.9306	0.9694	0.9773	0.9619

Framework versions

• Transformers 4.52.4
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.1