distilroberta-base-CoLA

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

Loss: 0.4974
Matthews Correlation: 0.5678

Model description

Mostly as a litmus test to see how it fares vs. the textattack one (should be similar) & associated metrics:

{
    "epoch": 4.0,
    "eval_loss": 0.49744734168052673,
    "eval_matthews_correlation": 0.5678267214677118,
    "eval_runtime": 1.9223,
    "eval_samples": 1043,
    "eval_samples_per_second": 542.586,
    "eval_steps_per_second": 135.777
}

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: 6e-05
train_batch_size: 32
eval_batch_size: 4
seed: 32010
distributed_type: multi-GPU
gradient_accumulation_steps: 4
total_train_batch_size: 128
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.03
num_epochs: 4.0
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Matthews Correlation
0.4778	1.0	67	0.4630	0.5161
0.4356	2.0	134	0.4725	0.5287
0.2934	3.0	201	0.4974	0.5678
0.1998	4.0	268	0.5419	0.5584

Framework versions

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