ft-google-gemma-2b-it-qlora

This model is a fine-tuned version of google/gemma-2b-it on the None dataset. It achieves the following results on the evaluation set:

• Loss: 3.7958

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 10
• eval_batch_size: 10
• seed: 42
• gradient_accumulation_steps: 10
• total_train_batch_size: 100
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant
• num_epochs: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
0.0077	100.0	100	2.3782
0.0005	200.0	200	2.9807
0.0004	300.0	300	3.1002
0.0004	400.0	400	3.1932
0.0004	500.0	500	3.2895
0.0004	600.0	600	3.3658
0.0003	700.0	700	3.3978
0.0004	800.0	800	3.4260
0.0004	900.0	900	3.5341
0.0003	1000.0	1000	3.5190
0.0004	1100.0	1100	3.5536
0.0003	1200.0	1200	3.5967
0.0003	1300.0	1300	3.6020
0.0004	1400.0	1400	3.6300
0.0004	1500.0	1500	3.6133
0.0003	1600.0	1600	3.7128
0.0003	1700.0	1700	3.7430
0.0003	1800.0	1800	3.7682
0.0003	1900.0	1900	3.7548
0.0003	2000.0	2000	3.7958

Framework versions

• PEFT 0.9.0
• Transformers 4.38.2
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2