CoolFace
Modelpublic

rikeshVertex/paddleocr-nepali-stage2-hw-digital-neg-fullft

sourceHugging Faceupdated 4mo agoView on Hugging Face
0likes2downloads
Model Card

<!-- 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. -->

paddleocr-nepali-stage2-hw-digital-neg-fullft

This model is a fine-tuned version of rikeshVertex/paddleocr-nepali-stage1-merged on an unknown dataset. It achieves the following results on the evaluation set:

  • —Loss: 0.0665

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
  • —trainbatchsize: 6
  • —evalbatchsize: 6
  • —seed: 42
  • —gradientaccumulationsteps: 6
  • —totaltrainbatch_size: 36
  • —optimizer: Use OptimizerNames.PAGEDADAMW with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
  • —lrschedulertype: cosine
  • —lrschedulerwarmup_steps: 261
  • —training_steps: 2612

Training results

Training LossEpochStepValidation Loss
0.24760.22981500.4683
0.14410.45973000.2510
0.09690.68954500.1371
0.09260.91936000.1173
0.06951.14867500.1015
0.05971.37849000.0959
0.05941.608310500.0847
0.05411.838112000.0768
0.05292.067413500.0749
0.05032.297215000.0772
0.05562.527116500.0748
0.05082.756918000.0701
0.04832.986719500.0630
0.03783.216021000.0693
0.03733.445922500.0636
0.05303.675724000.0642
0.04963.905525500.0647
0.04524.026120.0665

Framework versions

  • —PEFT 0.19.1
  • —Transformers 5.7.0
  • —Pytorch 2.11.0+cu130
  • —Tokenizers 0.22.2