---
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-reward-scale-05
  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-reward-scale-05

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: 0.5099
- Rewards/chosen: -2.9128
- Rewards/rejected: -4.2868
- Rewards/accuracies: 0.7328
- Rewards/margins: 1.3740
- Logps/rejected: -674.3233
- Logps/chosen: -575.2830
- Logits/rejected: 2.9146
- Logits/chosen: 0.8634

## 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 |
|:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
| 0.6688        | 0.1147 | 50   | 0.6561          | -0.0264        | -0.1297          | 0.6767             | 0.1033          | -258.6208      | -286.6466    | -2.5069         | -2.6066       |
| 0.5822        | 0.2294 | 100  | 0.5813          | -0.9069        | -1.4729          | 0.6724             | 0.5660          | -392.9425      | -374.6986    | 0.4795          | -0.0936       |
| 0.5512        | 0.3440 | 150  | 0.5534          | -0.9841        | -1.8057          | 0.7284             | 0.8216          | -426.2166      | -382.4128    | 2.9841          | 1.9098        |
| 0.5364        | 0.4587 | 200  | 0.5368          | -1.4185        | -2.3616          | 0.7328             | 0.9430          | -481.8062      | -425.8600    | 3.8766          | 2.4303        |
| 0.5308        | 0.5734 | 250  | 0.5235          | -2.3552        | -3.5698          | 0.7284             | 1.2145          | -602.6266      | -519.5305    | 4.2409          | 2.4548        |
| 0.532         | 0.6881 | 300  | 0.5117          | -2.5133        | -3.7138          | 0.7198             | 1.2004          | -617.0262      | -535.3408    | 3.0154          | 1.1080        |
| 0.5064        | 0.8028 | 350  | 0.5117          | -2.6361        | -3.9100          | 0.7241             | 1.2739          | -636.6490      | -547.6198    | 2.6854          | 0.7127        |
| 0.5105        | 0.9174 | 400  | 0.5099          | -2.9128        | -4.2868          | 0.7328             | 1.3740          | -674.3233      | -575.2830    | 2.9146          | 0.8634        |


### Framework versions

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