xtremedistil-l12-h384-uncased-CoLA

This model is a fine-tuned version of microsoft/xtremedistil-l12-h384-uncased on the GLUE COLA dataset. It achieves the following results on the evaluation set:

• Loss: 0.4974
• Matthews Correlation: 0.5396

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: 0.0001
• train_batch_size: 128
• eval_batch_size: 16
• seed: 5559
• 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: 16.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Matthews Correlation
0.4822	1.0	67	0.5893	0.2621
0.4669	2.0	134	0.5811	0.3722
0.3077	3.0	201	0.6150	0.4383
0.2594	4.0	268	0.4974	0.5396
0.21	5.0	335	0.5594	0.5182
0.1526	6.0	402	0.5715	0.5150
0.1775	7.0	469	0.6637	0.5020
0.1681	8.0	536	0.6958	0.5131
0.124	9.0	603	0.7057	0.5154
0.1111	10.0	670	0.8173	0.5074
0.1332	11.0	737	0.8253	0.5260
0.0673	12.0	804	0.8086	0.5180
0.0512	13.0	871	0.8409	0.5128
0.0457	14.0	938	0.8760	0.4947
0.04	15.0	1005	0.8522	0.5103
0.0485	16.0	1072	0.8556	0.5076

Framework versions

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