levshechter/tibetan-code-switching-detector

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

tibetan-code-switching-detector

This model is a fine-tuned version of OMRIDRORI/mbert-tibetan-continual-unicode-240k on an unknown dataset. It achieves the following results on the evaluation set:

• —Loss: 0.7828
• —Accuracy: 0.8124
• —Proximity F1: 0.0772
• —Proximity Recall: 0.2920
• —Proximity Precision: 0.0457
• —Exact Matches: 0.7963
• —Missed Switches: 0.0556
• —False Switches: 14.7685
• —Matches At 1 Words: 0.0093
• —Matches At 2 Words: 0.0
• —Matches At 3 Words: 0.0
• —Matches At 4 Words: 0.0
• —Matches At 5 Words: 0.0093
• —Matches At 6 Words: 0.0
• —Matches At 7 Words: 0.0
• —Matches At 8 Words: 0.0
• —Matches At 9 Words: 0.0
• —Matches At 10 Words: 0.0

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
• —trainbatchsize: 16
• —evalbatchsize: 16
• —seed: 42
• —gradientaccumulationsteps: 2
• —totaltrainbatch_size: 32
• —optimizer: Use adamwtorch with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
• —lrschedulertype: linear
• —lrschedulerwarmup_steps: 1000
• —num_epochs: 10
• —mixedprecisiontraining: Native AMP

Training results

Framework versions

• —Transformers 4.46.3
• —Pytorch 2.4.1+cu121
• —Datasets 2.0.0
• —Tokenizers 0.20.3