llama3-8b-instruct-qlora-large

This model is a fine-tuned version of LoftQ/Meta-Llama-3-8B-Instruct-4bit-64rank on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.8530

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: 2e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss
2.3454	1.0	158	1.2439
2.1288	2.0	316	1.0900
2.0335	3.0	474	1.0394
1.9315	4.0	632	0.9995
1.804	5.0	790	0.9605
1.6583	6.0	948	0.9411
1.4994	7.0	1106	0.9283
1.3388	8.0	1264	0.9158
1.1894	9.0	1422	0.9103
1.0616	10.0	1580	0.9027
0.9461	11.0	1738	0.8963
0.8447	12.0	1896	0.8922
0.7575	13.0	2054	0.8887
0.6817	14.0	2212	0.8803
0.6192	15.0	2370	0.8761
0.5669	16.0	2528	0.8715
0.5196	17.0	2686	0.8719
0.479	18.0	2844	0.8683
0.4473	19.0	3002	0.8662
0.4202	20.0	3160	0.8624
0.397	21.0	3318	0.8590
0.377	22.0	3476	0.8573
0.3622	23.0	3634	0.8558
0.3514	24.0	3792	0.8548
0.3434	25.0	3950	0.8543
0.3349	26.0	4108	0.8541
0.332	27.0	4266	0.8538
0.328	28.0	4424	0.8541
0.3286	29.0	4582	0.8532
0.3279	30.0	4740	0.8530

Framework versions

• PEFT 0.10.0
• Transformers 4.40.0
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.19.1