zephyr-7b-dpo-full-magpi-high-margin-3-epochs

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

• Loss: 0.0113
• Rewards/chosen: -2.2663
• Rewards/rejected: -55.9932
• Rewards/accuracies: 0.9980
• Rewards/margins: 53.7269
• Logps/rejected: -6240.1069
• Logps/chosen: -593.6095
• Logits/rejected: -1.4041
• Logits/chosen: -2.6538

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: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.0005	0.4739	50	0.0041	-0.9968	-34.4909	0.9980	33.4941	-4089.8755	-466.6592	-2.5370	-2.8673
0.0001	0.9479	100	0.0062	-2.2829	-48.3500	0.9980	46.0672	-5475.7905	-595.2667	-1.7708	-2.7952
0.0	1.4218	150	0.0108	-2.2233	-53.6662	0.9980	51.4429	-6007.4062	-589.3127	-1.6021	-2.7924
0.0	1.8957	200	0.0111	-2.2579	-55.3708	0.9980	53.1129	-6177.8711	-592.7751	-1.4403	-2.6854
0.0	2.3697	250	0.0113	-2.2670	-55.8932	0.9980	53.6262	-6230.1123	-593.6860	-1.4092	-2.6538
0.0	2.8436	300	0.0113	-2.2663	-55.9932	0.9980	53.7269	-6240.1069	-593.6095	-1.4041	-2.6538

Framework versions

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