zephyr-7b-dpop-uf6k-qlora-5e-7-epoch3

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the generation/UF6konly dataset. It achieves the following results on the evaluation set:

• Loss: 0.6877
• Positive Losses: 0.0653
• Dpo Losses: 0.6792
• Rewards/chosen: 0.0804
• Rewards/rejected: 0.0515
• Rewards/accuracies: 0.6786
• Rewards/margins: 0.0289
• Rewards/margins Max: 0.0914
• Rewards/margins Min: -0.0315
• Rewards/margins Std: 0.0551
• Logps/rejected: -254.0317
• Logps/chosen: -277.1849
• Logits/rejected: -2.8006
• Logits/chosen: -2.8458

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: 2
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• total_train_batch_size: 16
• 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: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Positive Losses	Dpo Losses	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
0.6932	0.3	100	0.6930	0.0052	0.6925	0.0096	0.0083	0.5794	0.0012	0.0066	-0.0040	0.0047	-258.3495	-284.2642	-2.8141	-2.8594
0.6919	0.61	200	0.6915	0.0119	0.6904	0.0242	0.0187	0.6667	0.0055	0.0195	-0.0083	0.0124	-257.3141	-282.7999	-2.8133	-2.8583
0.6903	0.91	300	0.6899	0.0165	0.6876	0.0395	0.0283	0.6667	0.0112	0.0379	-0.0143	0.0232	-256.3544	-281.2695	-2.8086	-2.8537
0.6832	1.22	400	0.6892	0.0304	0.6847	0.0525	0.0351	0.7024	0.0174	0.0557	-0.0196	0.0337	-255.6741	-279.9755	-2.8057	-2.8507
0.6776	1.52	500	0.6884	0.0444	0.6825	0.0647	0.0427	0.6905	0.0220	0.0710	-0.0256	0.0433	-254.9144	-278.7508	-2.8047	-2.8495
0.677	1.82	600	0.6873	0.0459	0.6811	0.0769	0.0519	0.6825	0.0250	0.0803	-0.0280	0.0484	-253.9932	-277.5360	-2.8047	-2.8494
0.6796	2.13	700	0.6872	0.0548	0.6800	0.0798	0.0526	0.6825	0.0272	0.0865	-0.0298	0.0521	-253.9202	-277.2366	-2.8026	-2.8477
0.6778	2.43	800	0.6875	0.0604	0.6795	0.0800	0.0518	0.6825	0.0282	0.0897	-0.0307	0.0540	-254.0074	-277.2222	-2.8024	-2.8474
0.6739	2.74	900	0.6878	0.0651	0.6793	0.0802	0.0515	0.6706	0.0287	0.0914	-0.0317	0.0550	-254.0345	-277.2037	-2.8028	-2.8477

Framework versions

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