zephyr-7b-dpo-full

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 0.3160
• Rewards/chosen: -4.1121
• Rewards/rejected: -8.3353
• Rewards/accuracies: 0.8201
• Rewards/margins: 4.2232
• Logps/rejected: -1123.6224
• Logps/chosen: -702.1752
• Logits/rejected: 0.5403
• Logits/chosen: -0.4404

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: 42
• distributed_type: multi-GPU
• num_devices: 4
• 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	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.5381	0.1152	100	0.4758	-1.9882	-2.9171	0.7270	0.9288	-581.7981	-489.7893	-2.8822	-2.9045
0.4268	0.2303	200	0.3577	-3.9068	-6.8487	0.7976	2.9419	-974.9606	-681.6494	-0.6781	-0.9791
0.4067	0.3455	300	0.3411	-3.9757	-7.6481	0.8094	3.6724	-1054.9027	-688.5351	-0.6642	-1.2474
0.4011	0.4607	400	0.3295	-4.4449	-8.4011	0.8156	3.9562	-1130.1991	-735.4550	0.1183	-0.7429
0.3727	0.5759	500	0.3260	-3.7203	-7.6540	0.8161	3.9337	-1055.4913	-662.9987	-0.4066	-1.3009
0.3933	0.6910	600	0.3190	-3.7331	-7.5182	0.8257	3.7851	-1041.9088	-664.2776	0.3247	-0.5819
0.3858	0.8062	700	0.3166	-3.9569	-8.0356	0.8246	4.0787	-1093.6547	-686.6614	0.3586	-0.6058
0.3785	0.9214	800	0.3161	-4.1174	-8.3387	0.8212	4.2213	-1123.9625	-702.7068	0.5558	-0.4246

Framework versions

• Transformers 4.44.1
• Pytorch 2.1.2+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1