yujiepan/test.mobilebert-uncased-12blks-squadv1-int8-f1-90.31

This model is a fine-tuned version of csarron/mobilebert-uncased-squad-v1 on the squad dataset with QAT.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 6e-05
• train_batch_size: 32
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine_with_restarts
• num_epochs: 15.0
• mixed_precision_training: Native AMP

Training results

torch model:

{
    "epoch": 14.82,
    "eval_exact_match": 83.66130558183538,
    "eval_f1": 90.30829859300735,
    "step": 41000
}

Framework versions

• Transformers 4.26.0
• Pytorch 1.13.1+cu116
• Datasets 2.8.0
• Tokenizers 0.13.2
• Optimum-intel 1.7.0