Llama0-3-8b-v0.1-dpo-lr6e-7-e1

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

• Loss: 0.6582
• Rewards/chosen: -0.7111
• Rewards/rejected: -0.8015
• Rewards/accuracies: 0.5927
• Rewards/margins: 0.0904
• Logps/rejected: -166.9069
• Logps/chosen: -159.4064
• Logits/rejected: 0.2042
• Logits/chosen: 0.1905

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: 6e-07
• train_batch_size: 2
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 8
• total_train_batch_size: 128
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6871	0.2137	100	0.6864	-0.0646	-0.0752	0.5927	0.0106	-94.2750	-94.7507	0.0781	0.0585
0.6735	0.4275	200	0.6722	-0.2864	-0.3235	0.6048	0.0371	-119.1064	-116.9293	0.1352	0.1179
0.6616	0.6412	300	0.6630	-0.5661	-0.6349	0.5968	0.0688	-150.2457	-144.8998	0.2044	0.1898
0.6599	0.8549	400	0.6591	-0.6785	-0.7632	0.5847	0.0846	-163.0742	-156.1471	0.2023	0.1883

Framework versions

• Transformers 4.45.1
• Pytorch 2.4.1+cu121
• Datasets 3.0.0
• Tokenizers 0.20.0