Edit model card

pixel-base-finetuned-qnli

This model is a fine-tuned version of Team-PIXEL/pixel-base on the GLUE QNLI dataset. It achieves the following results on the evaluation set:

  • Loss: 0.9503
  • Accuracy: 0.8860

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: 3e-05
  • train_batch_size: 64
  • eval_batch_size: 8
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 100
  • training_steps: 15000
  • mixed_precision_training: Apex, opt level O1

Training results

Training Loss Epoch Step Validation Loss Accuracy
0.5451 0.31 500 0.5379 0.7282
0.4451 0.61 1000 0.3846 0.8318
0.4567 0.92 1500 0.3543 0.8525
0.3558 1.22 2000 0.3294 0.8638
0.3324 1.53 2500 0.3221 0.8666
0.3434 1.83 3000 0.2976 0.8774
0.2573 2.14 3500 0.3193 0.8750
0.2411 2.44 4000 0.3044 0.8794
0.253 2.75 4500 0.2932 0.8834
0.1653 3.05 5000 0.3364 0.8841
0.1662 3.36 5500 0.3348 0.8797
0.1816 3.67 6000 0.3440 0.8869
0.1699 3.97 6500 0.3453 0.8845
0.1027 4.28 7000 0.4277 0.8810
0.0987 4.58 7500 0.4590 0.8832
0.0974 4.89 8000 0.4311 0.8783
0.0669 5.19 8500 0.5214 0.8819
0.0583 5.5 9000 0.5776 0.8850
0.065 5.8 9500 0.5646 0.8821
0.0381 6.11 10000 0.6252 0.8796
0.0314 6.41 10500 0.7222 0.8801
0.0453 6.72 11000 0.6951 0.8823
0.0264 7.03 11500 0.7620 0.8828
0.0215 7.33 12000 0.8160 0.8834
0.0176 7.64 12500 0.8583 0.8828
0.0245 7.94 13000 0.8484 0.8867
0.0124 8.25 13500 0.8927 0.8836
0.0112 8.55 14000 0.9368 0.8827
0.0154 8.86 14500 0.9405 0.8860
0.0046 9.16 15000 0.9503 0.8860

Framework versions

  • Transformers 4.17.0
  • Pytorch 1.11.0+cu113
  • Datasets 2.0.0
  • Tokenizers 0.11.6
Downloads last month
9
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Dataset used to train Team-PIXEL/pixel-base-finetuned-qnli

Evaluation results