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: -0.6585
Rewards/chosen: -1.5839
Rewards/rejected: -2.6984
Rewards/accuracies: 0.7857
Rewards/margins: 1.1145
Logps/rejected: -530.0449
Logps/chosen: -440.3677
Logits/rejected: 1.9295
Logits/chosen: 0.5950

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: 2
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	Logits/chosen	Logits/rejected	Logps/chosen	Logps/rejected	Validation Loss	Rewards/accuracies	Rewards/chosen	Rewards/margins	Rewards/rejected
0.9314	0.1047	100	-2.4608	-2.3941	-295.3837	-296.4655	0.9248	0.7103	-0.1341	0.2286	-0.3626
0.8722	0.2093	200	-1.9828	-1.6165	-334.3411	-367.4581	0.8865	0.7520	-0.5236	0.5489	-1.0726
0.8208	0.3140	300	-0.4089	0.3061	-370.6316	-428.0927	0.8215	0.7639	-0.8865	0.7924	-1.6789
0.8208	0.4186	400	-0.4262	0.5905	-401.0516	-460.0637	0.7982	0.7718	-1.1907	0.8079	-1.9986
0.7826	0.5233	500	1.0156	2.2339	-421.7270	-504.0349	0.7799	0.7758	-1.3975	1.0408	-2.4383
0.7546	0.6279	600	0.3290	1.6798	-437.6459	-526.8406	0.7723	0.7837	-1.5567	1.1097	-2.6664
0.7533	0.7326	700	0.6982	2.0190	-444.4420	-531.2306	0.7732	0.7837	-1.6247	1.0856	-2.7103
0.7498	0.8373	800	0.4246	1.7010	-437.6152	-523.4053	0.7710	0.7857	-1.5564	1.0756	-2.6320
0.7471	0.9419	900	0.5837	1.9213	-439.8712	-529.8352	0.7707	0.7857	-1.5789	1.1174	-2.6963

Framework versions

Transformers 4.44.2
Pytorch 2.2.1+cu118
Datasets 2.14.7
Tokenizers 0.19.1

YeongminKim/zephyr-7b-dpo-full

zephyr-7b-dpo-full

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

YeongminKim/
zephyr-7b-dpo-full