zephyr-7b-dpo-full-gpt_consistent-reward-scale-1-rpo

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0087
• Rewards/chosen: -0.0010
• Rewards/rejected: -0.1538
• Rewards/accuracies: 0.7759
• Rewards/margins: 0.1527
• Logps/rejected: -261.8980
• Logps/chosen: -285.1917
• Logits/rejected: -2.3893
• Logits/chosen: -2.4834

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: 8
• eval_batch_size: 8
• seed: 55
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 2
• 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: 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	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.012	0.1147	50	0.0113	0.0776	-0.0073	0.6983	0.0849	-247.2525	-277.3343	-2.5025	-2.5786
0.0111	0.2294	100	0.0100	0.0400	-0.0817	0.7112	0.1217	-254.6882	-281.0889	-2.3452	-2.4456
0.0104	0.3440	150	0.0098	-0.0092	-0.1421	0.7284	0.1329	-260.7338	-286.0115	-2.4006	-2.4971
0.0096	0.4587	200	0.0093	0.0230	-0.1186	0.7888	0.1416	-258.3851	-282.7939	-2.4206	-2.5115
0.0093	0.5734	250	0.0089	-0.0116	-0.1682	0.7845	0.1565	-263.3386	-286.2548	-2.3653	-2.4591
0.0096	0.6881	300	0.0088	-0.0083	-0.1589	0.7845	0.1506	-262.4115	-285.9173	-2.3891	-2.4814
0.0096	0.8028	350	0.0087	-0.0041	-0.1596	0.7802	0.1555	-262.4817	-285.5014	-2.3906	-2.4846
0.0093	0.9174	400	0.0087	-0.0010	-0.1538	0.7759	0.1527	-261.8980	-285.1917	-2.3893	-2.4834

Framework versions

• Transformers 4.44.0.dev0
• Pytorch 2.1.2
• Datasets 2.20.0
• Tokenizers 0.19.1