NoManDeRY/DPO-Shift-Qwen-2-7B-Ultrafeedback-fixed-0.95

qwen-2-7b-dpo-ultrafeedback-5e-7-SFTed-paged_adamw_32bit-fixed-0.95

This is a model released from the preprint: DPO-Shift: Shifting the Distribution of Direct Preference Optimization. Please refer to our repository for more details.

This model is a fine-tuned version of NoManDeRY/DPO-Shift-Qwen-2-7B-UltraChat200K-SFT on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 0.5890
• Rewards/chosen: -0.0170
• Rewards/rejected: -0.3976
• Dpo Lambda: 0.9500
• Rewards/accuracies: 0.7302
• Rewards/margins: 0.3806
• Logps/rejected: -346.4959
• Logps/chosen: -335.0127
• Logits/rejected: -1.2190
• Logits/chosen: -1.1200

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: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• 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	Dpo Lambda	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6842	0.1047	50	0.6827	0.1496	0.1349	0.9500	0.6865	0.0147	-293.2452	-318.3515	-1.2886	-1.1643
0.6498	0.2093	100	0.6609	0.2769	0.2144	0.9500	0.7381	0.0625	-285.2926	-305.6219	-1.2589	-1.1266
0.6549	0.3140	150	0.6408	0.2288	0.0982	0.9500	0.7341	0.1307	-296.9194	-310.4279	-1.3148	-1.1846
0.6413	0.4186	200	0.6250	0.1619	-0.0318	0.9500	0.7381	0.1938	-309.9195	-317.1192	-1.2956	-1.1761
0.6069	0.5233	250	0.6114	0.0886	-0.1684	0.9500	0.7302	0.2570	-323.5783	-324.4538	-1.2827	-1.1695
0.611	0.6279	300	0.5997	0.0461	-0.2674	0.9500	0.7381	0.3135	-333.4765	-328.6992	-1.2575	-1.1528
0.6151	0.7326	350	0.5924	-0.0016	-0.3586	0.9500	0.7222	0.3570	-342.5963	-333.4674	-1.2391	-1.1370
0.5997	0.8373	400	0.5898	-0.0127	-0.3884	0.9500	0.7222	0.3758	-345.5813	-334.5772	-1.2248	-1.1256
0.5708	0.9419	450	0.5890	-0.0170	-0.3976	0.9500	0.7302	0.3806	-346.4959	-335.0127	-1.2190	-1.1200

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.0+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1