zephyr-dpo-qlora-uf-oursuf6k-5e-7

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the generation/UF6k dataset. It achieves the following results on the evaluation set:

• Loss: 0.6794
• Rewards/chosen: 0.0105
• Rewards/rejected: -0.0268
• Rewards/accuracies: 0.6210
• Rewards/margins: 0.0373
• Rewards/margins Max: 0.1949
• Rewards/margins Min: -0.0994
• Rewards/margins Std: 0.0979
• Logps/rejected: -261.2557
• Logps/chosen: -283.5446
• Logits/rejected: -2.7323
• Logits/chosen: -2.7692

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-07
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Rewards/margins Max	Rewards/margins Min	Rewards/margins Std	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6876	0.15	100	0.6917	0.0050	0.0011	0.6080	0.0039	0.0235	-0.0125	0.0119	-258.4697	-284.0944	-2.7664	-2.8052
0.6792	0.29	200	0.6883	0.0130	-0.0001	0.6370	0.0131	0.0719	-0.0363	0.0359	-258.5862	-283.2935	-2.7570	-2.7954
0.6697	0.44	300	0.6849	0.0203	-0.0017	0.6170	0.0220	0.1184	-0.0597	0.0593	-258.7473	-282.5646	-2.7489	-2.7863
0.6571	0.58	400	0.6819	0.0205	-0.0099	0.6330	0.0303	0.1598	-0.0807	0.0801	-259.5654	-282.5452	-2.7352	-2.7727
0.6508	0.73	500	0.6802	0.0128	-0.0223	0.6270	0.0351	0.1844	-0.0939	0.0926	-260.8111	-283.3109	-2.7314	-2.7685
0.6444	0.88	600	0.6796	0.0113	-0.0257	0.6230	0.0369	0.1937	-0.0983	0.0973	-261.1460	-283.4656	-2.7308	-2.7678

Framework versions

• PEFT 0.7.1
• Transformers 4.39.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.14.6
• Tokenizers 0.15.2