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: -0.2185
Rewards/chosen: -1.3679
Rewards/rejected: -1.6737
Rewards/accuracies: 0.6925
Rewards/margins: 0.3058
Logps/rejected: -427.5733
Logps/chosen: -418.7628
Logits/rejected: -0.7419
Logits/chosen: -0.9285

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
2198.5145	0.2093	100	-0.4523	-0.7537	-1.2088	0.7183	0.4551	-381.0817	-357.3475	-1.1237	-1.2944
303.2607	0.4186	200	-0.3748	-1.3393	-1.6437	0.6825	0.3044	-424.5691	-415.9042	-0.7163	-0.9049
-0.2433	0.6279	300	-0.2201	-1.3689	-1.6741	0.6925	0.3052	-427.6178	-418.8684	-0.7414	-0.9283
-0.4102	0.8373	400	-0.2209	-1.3681	-1.6742	0.6905	0.3062	-427.6268	-418.7830	-0.7433	-0.9307

Framework versions

Transformers 4.44.2
Pytorch 2.2.1+cu118
Datasets 2.14.7
Tokenizers 0.19.1

YeongminKim/zephyr-7b-dpo-full-alpha0.5_batch128

zephyr-7b-dpo-full-alpha0.5_batch128

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

YeongminKim/
zephyr-7b-dpo-full-alpha0.5_batch128