metadata

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

zephyr-7b

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: 0.4399
Rewards/chosen: -3.1655
Rewards/rejected: -5.0200
Rewards/accuracies: 0.3398
Rewards/margins: 1.8544
Logps/rejected: -579.1896
Logps/chosen: -380.4651
Logits/rejected: 4.5948
Logits/chosen: 4.4604
Use Label: 6575.7188
Pred Label: 2212.2812

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: 4
eval_batch_size: 8
seed: 42
distributed_type: multi-GPU
num_devices: 8
gradient_accumulation_steps: 4
total_train_batch_size: 128
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	Use Label	Pred Label
0.6535	0.21	100	0.6432	-0.2049	-0.3593	0.3516	0.1544	-113.1259	-84.4063	-2.0537	-2.0656	1713.5	18.5
0.507	0.42	200	0.5048	-1.6723	-2.3466	0.3594	0.6743	-311.8494	-231.1388	2.1626	2.0915	3214.5625	373.4375
0.4799	0.63	300	0.4885	-1.7906	-2.6624	0.3359	0.8718	-343.4285	-242.9698	3.2225	3.1511	4474.75	969.25
0.4443	0.84	400	0.4405	-3.0809	-4.8915	0.3438	1.8106	-566.3419	-371.9976	4.5207	4.3874	5649.7188	1650.2812

Framework versions

PEFT 0.7.1
Transformers 4.38.2
Pytorch 2.1.1+cu121
Datasets 2.14.6
Tokenizers 0.15.2