HGU_rulebook-Llama3.2-Bllossom-5B_fine-tuning-QLoRA-8_16_5

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: 5.6999

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: 1e-05
• 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: 1570
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
12.6394	0.4964	78	12.3957
9.2796	0.9928	156	8.8049
7.1727	1.4893	234	7.0101
6.2211	1.9857	312	6.1729
5.9124	2.4821	390	5.9015
5.8102	2.9785	468	5.8015
5.7634	3.4749	546	5.7552
5.7337	3.9714	624	5.7330
5.7222	4.4678	702	5.7211
5.7148	4.9642	780	5.7148
5.7104	5.4606	858	5.7101
5.7054	5.9570	936	5.7066
5.7041	6.4535	1014	5.7043
5.6986	6.9499	1092	5.7024
5.7033	7.4463	1170	5.7014
5.7003	7.9427	1248	5.7006
5.7	8.4391	1326	5.7001
5.6966	8.9356	1404	5.6999
5.6971	9.4320	1482	5.6998
5.6974	9.9284	1560	5.6999

Framework versions

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