bert-trainer-8b

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

• Loss: 3.1639

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: 0.0005
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 1000
• num_epochs: 32
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
6.5416	1.0	500	6.5207
6.393	1.99	1000	6.3903
6.2817	2.99	1500	6.3033
6.2274	3.98	2000	6.2671
6.179	4.98	2500	6.2431
6.1684	5.98	3000	6.2309
6.1244	6.97	3500	6.2114
6.0879	7.97	4000	6.1932
6.0643	8.96	4500	6.1791
6.0481	9.96	5000	6.1638
6.0231	10.96	5500	6.1581
5.9987	11.95	6000	6.1365
5.9989	12.95	6500	6.1194
5.9535	13.94	7000	6.1095
5.9139	14.94	7500	6.0890
5.8462	15.94	8000	6.0224
5.7689	16.93	8500	5.9266
5.6137	17.93	9000	5.7195
4.7163	18.92	9500	4.6131
4.0877	19.92	10000	4.0903
3.7832	20.92	10500	3.8340
3.6104	21.91	11000	3.6572
3.4615	22.91	11500	3.5278
3.3661	23.9	12000	3.4201
3.271	24.9	12500	3.3333
3.2179	25.9	13000	3.2720
3.1759	26.89	13500	3.2317
3.1419	27.89	14000	3.2006
3.1041	28.88	14500	3.1806
3.0836	29.88	15000	3.1693
3.0998	30.88	15500	3.1679
3.08	31.87	16000	3.1639

Framework versions

• Transformers 4.26.1
• Pytorch 1.13.1
• Datasets 2.9.0
• Tokenizers 0.13.2