---
license: apache-2.0
base_model: alignment-handbook/zephyr-7b-sft-full
tags:
- trl
- dpo
- 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: 0.0327
- Rewards/chosen: -0.1258
- Rewards/rejected: -0.4253
- Rewards/accuracies: 0.7543
- Rewards/margins: 0.2995
- Logps/rejected: -289.0509
- Logps/chosen: -297.6692
- Logits/rejected: -1.4762
- Logits/chosen: -1.6863

## 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 |
|:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
| 0.0487        | 0.1147 | 50   | 0.0455          | 0.0300         | -0.0845          | 0.7026             | 0.1145          | -254.9693      | -282.0912    | -2.4928         | -2.5717       |
| 0.0408        | 0.2294 | 100  | 0.0391          | -0.1071        | -0.3089          | 0.6897             | 0.2018          | -277.4089      | -295.8004    | -1.7416         | -1.8404       |
| 0.0379        | 0.3440 | 150  | 0.0365          | -0.1406        | -0.4136          | 0.7155             | 0.2730          | -287.8874      | -299.1519    | -1.6357         | -1.7904       |
| 0.0365        | 0.4587 | 200  | 0.0350          | -0.0650        | -0.3264          | 0.7543             | 0.2614          | -279.1631      | -291.5866    | -1.7552         | -1.9178       |
| 0.0346        | 0.5734 | 250  | 0.0337          | -0.1319        | -0.4539          | 0.7543             | 0.3220          | -291.9156      | -298.2828    | -1.4871         | -1.7192       |
| 0.036         | 0.6881 | 300  | 0.0331          | -0.1336        | -0.4291          | 0.75               | 0.2955          | -289.4286      | -298.4504    | -1.4842         | -1.6835       |
| 0.0359        | 0.8028 | 350  | 0.0327          | -0.1378        | -0.4472          | 0.7586             | 0.3094          | -291.2399      | -298.8666    | -1.4658         | -1.6786       |
| 0.0351        | 0.9174 | 400  | 0.0327          | -0.1258        | -0.4253          | 0.7543             | 0.2995          | -289.0509      | -297.6692    | -1.4762         | -1.6863       |


### Framework versions

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