---
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: []
---

<!-- 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

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 HuggingFaceH4/ultrafeedback_binarized dataset.
It achieves the following results on the evaluation set:
- Loss: 0.5283
- Rewards/chosen: -0.0163
- Rewards/rejected: -1.2467
- Rewards/accuracies: 0.7738
- Rewards/margins: 1.2304
- Logps/rejected: -272.6863
- Logps/chosen: -282.1169
- Logits/rejected: -2.5360
- Logits/chosen: -2.5900

## 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: 4
- gradient_accumulation_steps: 2
- total_train_batch_size: 64
- total_eval_batch_size: 32
- 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 |
|:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
| 0.5446        | 0.1047 | 100  | 0.5753          | 1.0111         | 0.3529           | 0.7242             | 0.6581          | -256.6898      | -271.8434    | -2.5161         | -2.5743       |
| 0.5475        | 0.2093 | 200  | 0.5464          | 0.4347         | -0.4824          | 0.7639             | 0.9172          | -265.0432      | -277.6068    | -2.5380         | -2.5923       |
| 0.5359        | 0.3140 | 300  | 0.5473          | 0.0697         | -1.0170          | 0.7579             | 1.0867          | -270.3889      | -281.2571    | -2.5066         | -2.5596       |
| 0.5228        | 0.4186 | 400  | 0.5321          | -0.2311        | -1.3065          | 0.7540             | 1.0754          | -273.2837      | -284.2652    | -2.5933         | -2.6471       |
| 0.5217        | 0.5233 | 500  | 0.5260          | 0.0143         | -1.2073          | 0.7877             | 1.2216          | -272.2919      | -281.8111    | -2.5195         | -2.5773       |
| 0.517         | 0.6279 | 600  | 0.5262          | -0.2922        | -1.4562          | 0.7698             | 1.1640          | -274.7808      | -284.8755    | -2.5183         | -2.5744       |
| 0.4766        | 0.7326 | 700  | 0.5279          | -0.0183        | -1.2936          | 0.7798             | 1.2753          | -273.1544      | -282.1366    | -2.5194         | -2.5751       |
| 0.4894        | 0.8373 | 800  | 0.5257          | -0.0567        | -1.2594          | 0.7778             | 1.2027          | -272.8127      | -282.5211    | -2.5311         | -2.5851       |
| 0.4722        | 0.9419 | 900  | 0.5280          | -0.0160        | -1.2503          | 0.7798             | 1.2343          | -272.7223      | -282.1141    | -2.5362         | -2.5901       |


### Framework versions

- Transformers 4.40.2
- Pytorch 2.1.2
- Datasets 2.19.1
- Tokenizers 0.19.1