Modelpublic

ds3001/qlora-llama3b-iterative

sourceHugging Facellama3.2updated 2y agoView on Hugging Face

0likes9downloads

Model Card

qlora-llama3b-iterative

This model is a fine-tuned version of meta-llama/Llama-3.2-3B-Instruct on the train-iterative dataset. It achieves the following results on the evaluation set:

—Loss: 0.0051

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
—trainbatchsize: 1
—evalbatchsize: 1
—seed: 42
—gradientaccumulationsteps: 8
—totaltrainbatch_size: 8
—optimizer: Use OptimizerNames.ADAMWTORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizerargs=No additional optimizer arguments
—lrschedulertype: cosine
—lrschedulerwarmup_ratio: 0.1
—training_steps: 500

Training results

Training Loss	Epoch	Step	Validation Loss
2.1156	0.0889	10	1.5894
1.1893	0.1778	20	0.6868
0.5218	0.2667	30	0.4555
0.5292	0.3556	40	0.3795
0.3866	0.4444	50	0.3065
0.3232	0.5333	60	0.2074
0.1802	0.6222	70	0.1532
0.21	0.7111	80	0.1348
0.158	0.8	90	0.1372
0.1629	0.8889	100	0.1276
0.0966	0.9778	110	0.1003
0.0643	1.0667	120	0.0879
0.0726	1.1556	130	0.0872
0.0493	1.2444	140	0.0906
0.0746	1.3333	150	0.0587
0.0473	1.4222	160	0.0561
0.0644	1.5111	170	0.0503
0.0366	1.6	180	0.0307
0.0247	1.6889	190	0.0233
0.01	1.7778	200	0.0215
0.0393	1.8667	210	0.0122
0.0299	1.9556	220	0.0180
0.0166	2.0444	230	0.0082
0.0319	2.1333	240	0.0083
0.0077	2.2222	250	0.0072
0.0141	2.3111	260	0.0031
0.0017	2.4	270	0.0120
0.0015	2.4889	280	0.0153
0.0126	2.5778	290	0.0141
0.0043	2.6667	300	0.0022
0.0068	2.7556	310	0.0019
0.0018	2.8444	320	0.0022
0.0026	2.9333	330	0.0034
0.0017	3.0222	340	0.0076
0.0002	3.1111	350	0.0102
0.0004	3.2	360	0.0112
0.006	3.2889	370	0.0094
0.0003	3.3778	380	0.0075
0.0003	3.4667	390	0.0069
0.0002	3.5556	400	0.0067
0.0005	3.6444	410	0.0066
0.0003	3.7333	420	0.0072
0.0037	3.8222	430	0.0063
0.004	3.9111	440	0.0053
0.0003	4.0	450	0.0052
0.0002	4.0889	460	0.0051
0.0002	4.1778	470	0.0050
0.0006	4.2667	480	0.0049
0.0005	4.3556	490	0.0048
0.0002	4.4444	500	0.0051

Framework versions

—PEFT 0.12.0
—Transformers 4.46.1
—Pytorch 2.5.1+cu124
—Datasets 3.1.0
—Tokenizers 0.20.3