Mistral-7B-Instruct-v0.3-dpo-lora_lr1e-5_5ep

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

• Loss: 0.2423
• Rewards/chosen: -0.5640
• Rewards/rejected: -4.3641
• Rewards/accuracies: 0.8557
• Rewards/margins: 3.8002
• Logps/rejected: -417.8773
• Logps/chosen: -434.6749
• Logits/rejected: 0.0188
• Logits/chosen: 0.1716

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: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_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
0.4274	1.0	103	0.3324	0.2097	-1.9397	0.7932	2.1493	-393.6328	-426.9386	-0.3230	-0.0880
0.1309	2.0	206	0.2679	0.0296	-2.9700	0.8482	2.9997	-403.9364	-428.7391	-0.1288	0.0588
0.0376	3.0	309	0.2491	-0.4817	-4.1509	0.8452	3.6692	-415.7445	-433.8520	-0.0034	0.1539
0.0158	4.0	412	0.2450	-0.5678	-4.3787	0.8557	3.8110	-418.0231	-434.7127	0.0183	0.1715
0.0129	5.0	515	0.2423	-0.5640	-4.3641	0.8557	3.8002	-417.8773	-434.6749	0.0188	0.1716

Framework versions

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