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: 0.3329
Rewards/chosen: -3.5718
Rewards/rejected: -6.6220
Rewards/accuracies: 0.8281
Rewards/margins: 3.0502
Logps/rejected: -952.0885
Logps/chosen: -639.7113
Logits/rejected: -0.6186
Logits/chosen: -1.1189

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: 8
gradient_accumulation_steps: 4
total_train_batch_size: 256
total_eval_batch_size: 64
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
0.2094	0.2303	100	0.4273	-1.2973	-2.4710	0.7768	1.1737	-536.9892	-412.2621	-2.8747	-2.8886
0.1232	0.4606	200	0.3541	-2.8531	-5.1625	0.8069	2.3094	-806.1392	-567.8411	-1.2949	-1.6181
0.1	0.6908	300	0.3383	-3.9067	-6.9476	0.8158	3.0409	-984.6500	-673.2026	-0.4879	-0.9150
0.0956	0.9211	400	0.3329	-3.5971	-6.6557	0.8281	3.0586	-955.4623	-642.2396	-0.6052	-1.1035

Framework versions

Transformers 4.44.1
Pytorch 2.1.2+cu121
Datasets 2.21.0
Tokenizers 0.19.1

RichardErkhov/NicholasCorrado_-_zephyr-7b-group-dpo-full-4-4bits

Original model description:

zephyr-7b-group-dpo-full-4

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

RichardErkhov/
NicholasCorrado_-_zephyr-7b-group-dpo-full-4-4bits