HGU_rulebook-Llama3.2-Bllossom-5B_fine-tuning-QLoRA-64_16

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: 5.7052

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
12.8912	0.3182	50	12.8247
11.6977	0.6364	100	11.3812
9.1603	0.9547	150	8.9505
7.6735	1.2729	200	7.5549
6.7945	1.5911	250	6.7118
6.2617	1.9093	300	6.2235
6.0081	2.2275	350	5.9894
5.8829	2.5457	400	5.8750
5.8219	2.8640	450	5.8154
5.7859	3.1822	500	5.7831
5.7645	3.5004	550	5.7624
5.7485	3.8186	600	5.7478
5.7375	4.1368	650	5.7377
5.7339	4.4551	700	5.7301
5.7241	4.7733	750	5.7246
5.7212	5.0915	800	5.7204
5.7178	5.4097	850	5.7170
5.7158	5.7279	900	5.7145
5.7113	6.0461	950	5.7124
5.711	6.3644	1000	5.7107
5.7062	6.6826	1050	5.7093
5.7075	7.0008	1100	5.7082
5.7079	7.3190	1150	5.7074
5.7104	7.6372	1200	5.7067
5.7046	7.9554	1250	5.7063
5.7027	8.2737	1300	5.7058
5.7049	8.5919	1350	5.7056
5.7032	8.9101	1400	5.7053
5.7048	9.2283	1450	5.7053
5.7057	9.5465	1500	5.7052
5.7035	9.8648	1550	5.7052

Framework versions

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