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: 0.4929
Rewards/chosen: 21.1860
Rewards/rejected: 6.2518
Rewards/accuracies: 0.7344
Rewards/margins: 14.9342
Logps/rejected: -256.4154
Logps/chosen: -241.4075
Logits/rejected: -2.7091
Logits/chosen: -2.7366

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
0.5187	0.21	100	0.5296	19.0644	9.0310	0.7227	10.0334	-253.6362	-243.5290	-2.7384	-2.7638
0.508	0.42	200	0.5006	20.6504	7.0237	0.7266	13.6267	-255.6435	-241.9431	-2.7569	-2.7826
0.4808	0.63	300	0.4966	20.8183	6.9540	0.7227	13.8643	-255.7132	-241.7751	-2.7115	-2.7378
0.4835	0.84	400	0.4917	21.2230	6.3692	0.7344	14.8539	-256.2980	-241.3705	-2.7037	-2.7315

Framework versions

Transformers 4.38.2
Pytorch 2.1.2+cu118
Datasets 2.16.1
Tokenizers 0.15.2