zephyr-7b-dpo-full

This model was trained from scratch on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5107
• Rewards/chosen: -1.4645
• Rewards/rejected: -2.3555
• Rewards/accuracies: 0.7718
• Rewards/margins: 0.8911
• Logps/rejected: -491.4778
• Logps/chosen: -426.3907
• Logits/rejected: 1.4587
• Logits/chosen: 0.9514

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: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• total_eval_batch_size: 32
• 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
0.6339	0.1	100	0.6366	-0.4251	-0.6280	0.6766	0.2029	-318.7289	-322.4543	-1.7266	-1.8550
0.5801	0.21	200	0.5761	-0.9339	-1.4916	0.7242	0.5577	-405.0862	-373.3335	-1.7791	-1.8866
0.5298	0.31	300	0.5505	-0.9519	-1.6203	0.7401	0.6684	-417.9537	-375.1365	-0.9729	-1.1938
0.5055	0.42	400	0.5331	-1.3809	-2.1858	0.7540	0.8048	-474.5050	-418.0395	0.2901	-0.0376
0.5243	0.52	500	0.5240	-1.5398	-2.3578	0.7718	0.8180	-491.7054	-433.9210	1.1167	0.7245
0.5024	0.63	600	0.5212	-1.6677	-2.5319	0.75	0.8643	-509.1215	-446.7127	1.3224	0.8469
0.4855	0.73	700	0.5156	-1.5293	-2.4112	0.7579	0.8819	-497.0490	-432.8780	1.5165	1.0177
0.5048	0.84	800	0.5121	-1.4754	-2.3714	0.7698	0.8960	-493.0640	-427.4831	1.3869	0.8797
0.5193	0.94	900	0.5109	-1.4545	-2.3434	0.7738	0.8889	-490.2650	-425.3930	1.4499	0.9411

Framework versions

• Transformers 4.36.2
• Pytorch 2.1.2
• Datasets 2.14.6
• Tokenizers 0.15.0