This model is a fine-tuned version of google/electra-base-discriminator on the GLUE COLA dataset. It achieves the following results on the evaluation set:

Loss: 0.3542
Matthews Correlation: 0.6580

Model description

Trying to find a decent optimum between accuracy/quality and inference speed.

json

{
    "epoch": 8.0,
    "eval_loss": 0.3541961908340454,
    "eval_matthews_correlation": 0.6579677841732349,
    "eval_runtime": 1.9552,
    "eval_samples": 1043,
    "eval_samples_per_second": 533.451,
    "eval_steps_per_second": 33.756
}

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: 8e-05
train_batch_size: 128
eval_batch_size: 16
seed: 22165
distributed_type: multi-GPU
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.03
num_epochs: 8.0
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Matthews Correlation
0.4004	1.0	67	0.3569	0.6340
0.2843	2.0	134	0.3542	0.6580
0.1228	3.0	201	0.4201	0.6412
0.0989	4.0	268	0.4780	0.6757
0.0681	5.0	335	0.4900	0.6925
0.0506	6.0	402	0.5837	0.6785
0.0093	7.0	469	0.6298	0.6652
0.0244	8.0	536	0.6292	0.6750

Framework versions

Transformers 4.27.0.dev0
Pytorch 1.13.1+cu117
Datasets 2.8.0
Tokenizers 0.13.1

pszemraj/electra-base-discriminator-CoLA

electra-base-discriminator-CoLA