metadata

license: apache-2.0
library_name: peft
tags:
  - alignment-handbook
  - generated_from_trainer
  - trl
  - dpo
  - generated_from_trainer
datasets:
  - HuggingFaceH4/ultrafeedback_binarized
base_model: mistralai/Mistral-7B-v0.1
model-index:
  - name: zephyr-7b-dpo-qlora
    results: []

zephyr-7b-dpo-qlora

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

Loss: 2121.0452
Rewards/chosen: 0.0578
Rewards/rejected: -0.0912
Rewards/accuracies: 0.7599
Rewards/margins: 0.1490
Logps/rejected: -253.8891
Logps/chosen: -259.2458
Logits/rejected: -2.2028
Logits/chosen: -2.2552

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-06
train_batch_size: 2
eval_batch_size: 8
seed: 42
distributed_type: multi-GPU
num_devices: 4
gradient_accumulation_steps: 16
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
2149.4746	0.21	100	2190.7666	0.0445	-0.0848	0.7460	0.1293	-253.2523	-260.5782	-2.1770	-2.2229
2105.1256	0.42	200	2151.1555	0.0543	-0.0961	0.7599	0.1504	-254.3840	-259.5941	-2.2074	-2.2603
2135.4973	0.63	300	2129.0896	0.0626	-0.0799	0.7560	0.1425	-252.7585	-258.7624	-2.2232	-2.2765
2099.8018	0.84	400	2121.6672	0.0538	-0.0959	0.7540	0.1497	-254.3591	-259.6440	-2.2016	-2.2541

Framework versions

PEFT 0.7.1
Transformers 4.36.2
Pytorch 2.1.1+cu121
Datasets 2.14.6
Tokenizers 0.15.2