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

Loss: 0.4403
Matthews Correlation: 0.5510

Model description

trying to optimize accuracy/speed:

json

{
    "epoch": 8.0,
    "eval_loss": 0.4402828514575958,
    "eval_matthews_correlation": 0.5510400717227824,
    "eval_runtime": 0.9341,
    "eval_samples": 1043,
    "eval_samples_per_second": 1116.545,
    "eval_steps_per_second": 70.654
}

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: 512
eval_batch_size: 16
seed: 32754
distributed_type: multi-GPU
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.05
num_epochs: 8.0
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Matthews Correlation
0.6139	1.0	17	0.5997	0.0
0.5315	2.0	34	0.4890	0.5154
0.4244	3.0	51	0.4469	0.5433
0.3568	4.0	68	0.4403	0.5510
0.319	5.0	85	0.4517	0.5654
0.2887	6.0	102	0.4656	0.5728
0.2771	7.0	119	0.4558	0.5883
0.2729	8.0	136	0.4569	0.5858

Framework versions

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

pszemraj/electra-small-discriminator-CoLA

electra-small-discriminator-CoLA