Modelpublic

HachiML/Mistral-7B-v0.3-dpo-lora_sr_mxm1_lr1e-5_5ep

sourceHugging Faceupdated 2y agoView on Hugging Face

0likes13downloads

Model Card

Mistral-7B-v0.3-dpo-lorasrmxm1lr1e-55ep

This model is a fine-tuned version of HachiML/Mistral-7B-v0.3-mxm0.5-lora on an unknown dataset. It achieves the following results on the evaluation set:

—Loss: 0.2840
—Rewards/chosen: -2.3833
—Rewards/rejected: -5.5556
—Rewards/accuracies: 0.8854
—Rewards/margins: 3.1723
—Logps/rejected: -296.6647
—Logps/chosen: -395.2098
—Logits/rejected: 2.1386
—Logits/chosen: 1.2143

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
—trainbatchsize: 16
—evalbatchsize: 16
—seed: 42
—optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
—lrschedulertype: cosine
—lrschedulerwarmup_ratio: 0.05
—num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.471	1.0	103	0.3446	-0.9519	-2.6013	0.8304	1.6494	-267.1219	-380.8956	0.8179	0.3512
0.191	2.0	206	0.3087	-1.6824	-4.1087	0.8512	2.4263	-282.1961	-388.2008	1.3280	0.7210
0.0863	3.0	309	0.2806	-2.0077	-4.9634	0.9062	2.9558	-290.7433	-391.4536	1.7628	1.0070
0.0485	4.0	412	0.2810	-2.4278	-5.6118	0.8854	3.1840	-297.2274	-395.6552	2.1611	1.2277
0.0392	5.0	515	0.2840	-2.3833	-5.5556	0.8854	3.1723	-296.6647	-395.2098	2.1386	1.2143

Framework versions

—PEFT 0.11.1
—Transformers 4.41.1
—Pytorch 2.3.0+cu121
—Datasets 2.19.1
—Tokenizers 0.19.1