bert_uncased_L-2_H-256_A-4-mlm-multi-emails-hq

This model is a fine-tuned version of google/bert_uncased_L-2_H-256_A-4 on the postbot/multi-emails-hq dataset. It achieves the following results on the evaluation set:

• Loss: 2.4596
• Accuracy: 0.5642

Model description

This is a ~40MB version of BERT finetuned on an MLM task on email data.

Intended uses & limitations

• this is mostly a test/example

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• train_batch_size: 8
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 16
• 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: 0.05
• num_epochs: 8.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
3.097	0.99	141	2.8195	0.5180
2.9097	1.99	282	2.6704	0.5367
2.8335	2.99	423	2.5764	0.5485
2.7433	3.99	564	2.5213	0.5563
2.6828	4.99	705	2.4667	0.5641
2.666	5.99	846	2.4688	0.5642
2.6517	6.99	987	2.4452	0.5679
2.6309	7.99	1128	2.4596	0.5642

Framework versions

• Transformers 4.27.0.dev0
• Pytorch 2.0.0.dev20230129+cu118
• Datasets 2.8.0
• Tokenizers 0.13.1