HGU_rulebook-Llama3.2-Bllossom-5B_fine-tuning-QLoRA-8_32_6

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: 5.6808

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: 1884
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
5.9932	0.5982	94	5.8611
5.6904	1.1965	188	5.6907
5.6867	1.7947	282	5.6848
5.6817	2.3930	376	5.6827
5.6787	2.9912	470	5.6810
5.6761	3.5895	564	5.6809
5.6719	4.1877	658	5.6807
5.6775	4.7860	752	5.6802
5.6745	5.3842	846	5.6804
5.6768	5.9825	940	5.6797
5.6737	6.5807	1034	5.6799
5.6724	7.1790	1128	5.6802
5.6718	7.7772	1222	5.6800
5.6729	8.3755	1316	5.6801
5.6693	8.9737	1410	5.6801
5.6723	9.5720	1504	5.6806
5.6719	10.1702	1598	5.6806
5.6636	10.7685	1692	5.6808
5.6714	11.3667	1786	5.6808
5.671	11.9650	1880	5.6808

Framework versions

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