Llama-2-7b-hf-eval_threapist-DPO-filtered-0.2-local-version-1

This model is a fine-tuned version of meta-llama/Llama-2-7b-hf on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.8124
• Rewards/chosen: -0.4914
• Rewards/rejected: -0.6417
• Rewards/accuracies: 0.5
• Rewards/margins: 0.1503
• Logps/rejected: -52.4510
• Logps/chosen: -51.8579
• Logits/rejected: -1.2607
• Logits/chosen: -1.2496

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: 5e-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 4
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.4633	0.4	94	0.7283	-0.4834	-0.5276	0.5	0.0443	-51.3107	-51.7779	-0.9037	-0.8864
0.7207	0.8	188	0.7368	-0.2345	-0.2555	0.6000	0.0210	-48.5893	-49.2894	-0.7571	-0.7388
0.3511	1.2	282	0.7674	-0.4686	-0.6027	0.5500	0.1341	-52.0614	-51.6306	-1.0377	-1.0234
0.6543	1.6	376	0.8144	-0.4911	-0.6379	0.5	0.1468	-52.4135	-51.8557	-1.2458	-1.2342
0.123	2.0	470	0.8124	-0.4914	-0.6417	0.5	0.1503	-52.4510	-51.8579	-1.2607	-1.2496

Framework versions

• PEFT 0.10.0
• Transformers 4.40.2
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1