metadata

license: apache-2.0
base_model: alignment-handbook/zephyr-7b-sft-full
tags:
  - alignment-handbook
  - trl
  - dpo
  - generated_from_trainer
  - trl
  - dpo
  - generated_from_trainer
datasets:
  - HuggingFaceH4/ultrafeedback_binarized
model-index:
  - name: zephyr-7b-dpo-full
    results: []

zephyr-7b-dpo-full

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: 0.5352
Rewards/chosen: 0.5766
Rewards/rejected: -0.3207
Rewards/accuracies: 0.7617
Rewards/margins: 0.8972
Logps/rejected: -269.0807
Logps/chosen: -251.0624
Logits/rejected: -2.4374
Logits/chosen: -2.4784

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: 2
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
0.565	0.21	100	0.5718	0.5950	-0.0056	0.7383	0.6006	-262.7792	-250.6930	-2.5105	-2.5504
0.5467	0.42	200	0.5433	0.6478	-0.1115	0.7461	0.7594	-264.8979	-249.6371	-2.4783	-2.5179
0.517	0.63	300	0.5370	0.5686	-0.2689	0.7695	0.8374	-268.0445	-251.2220	-2.5203	-2.5623
0.518	0.84	400	0.5348	0.6286	-0.2212	0.7539	0.8498	-267.0915	-250.0218	-2.4324	-2.4731

Framework versions

Transformers 4.38.2
Pytorch 2.1.2+cu118
Datasets 2.16.1
Tokenizers 0.15.2