---
license: apache-2.0
base_model: alignment-handbook/zephyr-7b-sft-full
tags:
- trl
- dpo
- alignment-handbook
- generated_from_trainer
model-index:
- name: zephyr-7b-dpo-full-gpt_consistent-reward-scale-1-rpo
  results: []
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# zephyr-7b-dpo-full-gpt_consistent-reward-scale-1-rpo

This model is a fine-tuned version of [alignment-handbook/zephyr-7b-sft-full](https://huggingface.co/alignment-handbook/zephyr-7b-sft-full) on the None dataset.
It achieves the following results on the evaluation set:
- Loss: 0.1126
- Rewards/chosen: -0.4018
- Rewards/rejected: -0.9642
- Rewards/accuracies: 0.75
- Rewards/margins: 0.5623
- Logps/rejected: -342.9399
- Logps/chosen: -325.2745
- Logits/rejected: 0.1904
- Logits/chosen: -0.6159

## 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: 55
- 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 |
|:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
| 0.1732        | 0.1147 | 50   | 0.1640          | 0.0140         | -0.1117          | 0.7069             | 0.1256          | -257.6913      | -283.6945    | -2.4927         | -2.5723       |
| 0.1412        | 0.2294 | 100  | 0.1392          | -0.5601        | -0.9388          | 0.7198             | 0.3787          | -340.4027      | -341.0970    | -0.8338         | -1.1029       |
| 0.1327        | 0.3440 | 150  | 0.1257          | -0.3660        | -0.8686          | 0.7241             | 0.5025          | -333.3776      | -321.6920    | -0.4213         | -0.9346       |
| 0.1242        | 0.4587 | 200  | 0.1193          | -0.3424        | -0.8506          | 0.7457             | 0.5082          | -331.5807      | -319.3252    | -0.1498         | -0.9084       |
| 0.118         | 0.5734 | 250  | 0.1167          | -0.3734        | -0.9685          | 0.7543             | 0.5950          | -343.3683      | -322.4342    | -0.0251         | -0.8147       |
| 0.1236        | 0.6881 | 300  | 0.1145          | -0.4780        | -1.0449          | 0.7629             | 0.5669          | -351.0162      | -332.8936    | 0.2891          | -0.5285       |
| 0.122         | 0.8028 | 350  | 0.1127          | -0.4577        | -1.0376          | 0.7672             | 0.5799          | -350.2830      | -330.8570    | 0.3397          | -0.4806       |
| 0.12          | 0.9174 | 400  | 0.1126          | -0.4018        | -0.9642          | 0.75               | 0.5623          | -342.9399      | -325.2745    | 0.1904          | -0.6159       |


### Framework versions

- Transformers 4.44.0.dev0
- Pytorch 2.1.2
- Datasets 2.20.0
- Tokenizers 0.19.1