zephyr-7b-dpop-uf6k-qlora-5e-6-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: 0.9050
• Positive Losses: 1.9848
• Dpo Losses: 0.6341
• Rewards/chosen: 0.1274
• Rewards/rejected: -0.0080
• Rewards/accuracies: 0.7381
• Rewards/margins: 0.1354
• Rewards/margins Max: 0.3634
• Rewards/margins Min: -0.0877
• Rewards/margins Std: 0.2033
• Logps/rejected: -259.9846
• Logps/chosen: -272.4854
• Logits/rejected: -2.7772
• Logits/chosen: -2.8182

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-06
• 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
0.6895	0.3	100	0.6866	0.0402	0.6797	0.0926	0.0647	0.6905	0.0279	0.0904	-0.0340	0.0553	-252.7092	-275.9622	-2.7972	-2.8420
0.682	0.61	200	0.6814	0.0885	0.6688	0.1594	0.1077	0.7183	0.0518	0.1531	-0.0501	0.0916	-248.4167	-269.2792	-2.8159	-2.8598
0.672	0.91	300	0.6821	0.1534	0.6642	0.1744	0.1117	0.7262	0.0626	0.1928	-0.0674	0.1166	-248.0109	-267.7851	-2.7544	-2.7992
0.6084	1.22	400	0.7278	0.5292	0.6533	0.1649	0.0771	0.7183	0.0878	0.2556	-0.0760	0.1485	-251.4729	-268.7308	-2.7666	-2.8121
0.6009	1.52	500	0.7518	0.7332	0.6465	0.1548	0.0515	0.7341	0.1033	0.2829	-0.0732	0.1598	-254.0323	-269.7391	-2.7510	-2.7945
0.613	1.82	600	0.7997	1.0438	0.6410	0.1367	0.0201	0.7619	0.1166	0.3122	-0.0805	0.1767	-257.1772	-271.5526	-2.7849	-2.8271
0.5504	2.13	700	0.8485	1.5090	0.6371	0.1361	0.0091	0.7540	0.1271	0.3381	-0.0856	0.1905	-258.2727	-271.6062	-2.7839	-2.8261
0.5924	2.43	800	0.9247	2.1422	0.6323	0.1214	-0.0182	0.7341	0.1396	0.3676	-0.0900	0.2062	-261.0059	-273.0837	-2.7895	-2.8309
0.5608	2.74	900	0.9056	1.9910	0.6338	0.1271	-0.0088	0.7460	0.1359	0.3635	-0.0870	0.2030	-260.0656	-272.5118	-2.7790	-2.8200

Framework versions

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