zephyr-7b-dpo-full-magpi-low-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: 0.0010
• Rewards/chosen: -3.9690
• Rewards/rejected: -58.1646
• Rewards/accuracies: 0.9980
• Rewards/margins: 54.1956
• Logps/rejected: -6457.2515
• Logps/chosen: -763.8833
• Logits/rejected: 4.2613
• Logits/chosen: 1.2520

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
0.009	0.4739	50	0.0026	-0.6343	-34.9234	0.9960	34.2890	-4133.1240	-430.4127	-2.6406	-2.9323
0.0158	0.9479	100	0.0003	-1.2700	-43.4290	1.0	42.1589	-4983.6846	-493.9866	3.0564	-0.5346
0.0007	1.4218	150	0.0005	-2.9913	-52.6639	0.9980	49.6726	-5907.1768	-666.1114	4.1711	0.7924
0.0015	1.8957	200	0.0006	-3.5933	-54.2858	0.9980	50.6925	-6069.3657	-726.3069	4.0713	1.0487
0.0	2.3697	250	0.0009	-3.9372	-57.6402	0.9980	53.7030	-6404.8037	-760.6977	4.2488	1.2599
0.0001	2.8436	300	0.0010	-3.9690	-58.1646	0.9980	54.1956	-6457.2515	-763.8833	4.2613	1.2520

Framework versions

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