HGU_rulebook-Llama3.2-Bllossom-5B_fine-tuning-QLoRA-8_32_4

This model is a fine-tuned version of Bllossom/llama-3.2-Korean-Bllossom-AICA-5B on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 5.6787

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: 0.0002
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Use adamw_bnb_8bit with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine_with_restarts
• lr_scheduler_warmup_ratio: 0.1
• training_steps: 1256
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
6.8664	0.3946	62	6.3467
5.7104	0.7892	124	5.7018
5.6856	1.1838	186	5.6870
5.6807	1.5784	248	5.6836
5.6836	1.9730	310	5.6826
5.6797	2.3675	372	5.6812
5.6777	2.7621	434	5.6804
5.6783	3.1567	496	5.6804
5.6758	3.5513	558	5.6794
5.6759	3.9459	620	5.6793
5.675	4.3405	682	5.6794
5.676	4.7351	744	5.6788
5.6721	5.1297	806	5.6789
5.6737	5.5243	868	5.6787
5.6719	5.9189	930	5.6785
5.6751	6.3134	992	5.6786
5.6698	6.7080	1054	5.6787
5.6709	7.1026	1116	5.6787
5.6742	7.4972	1178	5.6787
5.6721	7.8918	1240	5.6787

Framework versions

• PEFT 0.12.0
• Transformers 4.46.2
• Pytorch 2.0.1+cu118
• Datasets 3.0.0
• Tokenizers 0.20.1