---
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: 0.1132
- Rewards/chosen: -0.3948
- Rewards/rejected: -0.9598
- Rewards/accuracies: 0.7543
- Rewards/margins: 0.5650
- Logps/rejected: -342.4998
- Logps/chosen: -324.5692
- Logits/rejected: 1.4353
- Logits/chosen: 0.4067

## 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   | 0.1641          | 0.0166         | -0.1092          | 0.7069             | 0.1259          | -257.4446      | -283.4267    | -2.4943         | -2.5737       |
| 0.1412        | 0.2294 | 100  | 0.1347          | -0.4342        | -0.7962          | 0.7284             | 0.3620          | -326.1406      | -328.5106    | -0.1538         | -0.4184       |
| 0.1307        | 0.3440 | 150  | 0.1261          | -0.3553        | -0.8583          | 0.7284             | 0.5030          | -332.3533      | -320.6210    | 0.7144          | 0.0181        |
| 0.1238        | 0.4587 | 200  | 0.1199          | -0.4108        | -0.9476          | 0.7328             | 0.5368          | -341.2862      | -326.1717    | 1.2989          | 0.2969        |
| 0.1185        | 0.5734 | 250  | 0.1166          | -0.3086        | -0.8924          | 0.7543             | 0.5838          | -335.7633      | -315.9550    | 0.8516          | -0.1745       |
| 0.1228        | 0.6881 | 300  | 0.1155          | -0.3695        | -0.9267          | 0.7457             | 0.5571          | -339.1875      | -322.0434    | 0.8574          | -0.1316       |
| 0.1213        | 0.8028 | 350  | 0.1136          | -0.4396        | -1.0157          | 0.7629             | 0.5762          | -348.0973      | -329.0486    | 1.5740          | 0.5152        |
| 0.12          | 0.9174 | 400  | 0.1132          | -0.3948        | -0.9598          | 0.7543             | 0.5650          | -342.4998      | -324.5692    | 1.4353          | 0.4067        |


### Framework versions

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