DisamBertCrossEncoder-base

This model is a fine-tuned version of answerdotai/ModernBERT-base on the semcor dataset. It achieves the following results on the evaluation set:

• Loss: 13.9274
• Precision: 0.6274
• Recall: 0.6398
• F1: 0.6335
• Matthews: 0.6392

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: 0.0001
• train_batch_size: 64
• 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: inverse_sqrt
• lr_scheduler_warmup_steps: 1000
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Matthews
No log	0	0	427.8458	0.5014	0.4790	0.4899	0.4781
9.0689	1.0	3504	15.5744	0.6010	0.6196	0.6102	0.6190
8.5420	2.0	7008	15.4129	0.6088	0.6253	0.6170	0.6247
7.8106	3.0	10512	14.3562	0.6138	0.6328	0.6232	0.6322
7.6303	4.0	14016	13.9741	0.6157	0.6372	0.6262	0.6366
7.6930	5.0	17520	13.8324	0.6262	0.6402	0.6331	0.6397
7.4897	6.0	21024	13.9649	0.6144	0.6323	0.6232	0.6318
7.3819	7.0	24528	13.4877	0.6273	0.6407	0.6339	0.6401
7.4083	8.0	28032	13.7249	0.6321	0.6402	0.6362	0.6397
7.0140	9.0	31536	13.5219	0.6168	0.6389	0.6277	0.6383
7.7287	10.0	35040	13.9274	0.6274	0.6398	0.6335	0.6392

Framework versions

• Transformers 5.2.0
• Pytorch 2.10.0+cu128
• Datasets 4.5.0
• Tokenizers 0.22.2