Edit model card

LiLT-SER-EN

This model is a fine-tuned version of nielsr/lilt-xlm-roberta-base on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 2.3960
  • Precision: 0.7248
  • Recall: 0.7458
  • F1: 0.7351
  • Accuracy: 0.7438

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: 5e-05
  • train_batch_size: 8
  • eval_batch_size: 2
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • training_steps: 10000

Training results

Training Loss Epoch Step Validation Loss Precision Recall F1 Accuracy
0.0832 6.67 500 1.0009 0.6741 0.6923 0.6831 0.7292
0.052 13.33 1000 1.4186 0.7225 0.7320 0.7272 0.7441
0.0027 20.0 1500 1.5508 0.7218 0.7376 0.7297 0.7464
0.0034 26.67 2000 1.7198 0.7051 0.7382 0.7213 0.7422
0.002 33.33 2500 1.8116 0.7106 0.7392 0.7246 0.7424
0.0002 40.0 3000 1.8843 0.6769 0.7514 0.7122 0.7435
0.0009 46.67 3500 1.9528 0.7401 0.7514 0.7457 0.7518
0.0224 53.33 4000 2.0602 0.7178 0.7529 0.7350 0.7476
0.0002 60.0 4500 2.2901 0.7283 0.7509 0.7394 0.7287
0.0001 66.67 5000 2.1746 0.7198 0.7433 0.7313 0.7371
0.0001 73.33 5500 1.9452 0.7214 0.7387 0.7299 0.7641
0.0 80.0 6000 2.0976 0.7350 0.7560 0.7454 0.7442
0.0021 86.67 6500 2.3034 0.7200 0.7387 0.7292 0.7365
0.0 93.33 7000 2.2409 0.7348 0.7636 0.7489 0.7499
0.0 100.0 7500 2.2742 0.7362 0.7193 0.7276 0.7472
0.0 106.67 8000 2.4953 0.7312 0.7509 0.7409 0.7363
0.0 113.33 8500 2.4936 0.7340 0.7438 0.7389 0.7396
0.0 120.0 9000 2.3976 0.7239 0.7453 0.7344 0.7440
0.0001 126.67 9500 2.3723 0.7282 0.7478 0.7379 0.7441
0.0 133.33 10000 2.3960 0.7248 0.7458 0.7351 0.7438

Framework versions

  • Transformers 4.35.2
  • Pytorch 2.1.0+cu121
  • Datasets 2.18.0
  • Tokenizers 0.15.1
Downloads last month
1
Safetensors
Model size
284M params
Tensor type
F32
·
Inference API
This model can be loaded on Inference API (serverless).

Finetuned from