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: 0.5018
Rewards/chosen: -1.1027
Rewards/rejected: -2.0673
Rewards/accuracies: 0.7656
Rewards/margins: 0.9646
Logps/rejected: -469.4237
Logps/chosen: -372.8636
Logits/rejected: -0.4162
Logits/chosen: -1.1470

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
0.5767	0.21	100	0.5762	-0.5418	-1.0929	0.7148	0.5512	-371.9847	-316.7739	-1.7636	-1.7624
0.5185	0.42	200	0.5312	-1.0541	-1.7947	0.7656	0.7407	-442.1638	-368.0010	-0.6928	-1.0192
0.5166	0.63	300	0.5114	-1.1529	-2.0654	0.7539	0.9125	-469.2292	-377.8804	-0.6568	-1.2977
0.5003	0.84	400	0.5023	-1.0875	-2.0687	0.7695	0.9812	-469.5586	-371.3432	-0.3666	-1.1044

Framework versions

Transformers 4.39.0.dev0
Pytorch 2.1.2
Datasets 2.18.1.dev0
Tokenizers 0.15.2