---
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-ultrabin-reward-scale-01
  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-ultrabin-reward-scale-01

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: 0.5340
- Rewards/chosen: -1.6749
- Rewards/rejected: -3.0354
- Rewards/accuracies: 0.7852
- Rewards/margins: 1.3605
- Logps/rejected: -566.2047
- Logps/chosen: -430.1231
- Logits/rejected: 2.3565
- Logits/chosen: 1.3978

## 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 |
|:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
| 0.6687        | 0.1046 | 50   | 0.6493          | 0.0258         | -0.0870          | 0.7070             | 0.1128          | -271.3634      | -260.0484    | -2.5783         | -2.6158       |
| 0.5614        | 0.2092 | 100  | 0.5807          | -0.8058        | -1.5920          | 0.7109             | 0.7862          | -421.8647      | -343.2120    | -0.2227         | -0.5290       |
| 0.5419        | 0.3138 | 150  | 0.5585          | -1.0477        | -2.0655          | 0.7461             | 1.0179          | -469.2165      | -367.3957    | 0.6415          | -0.0014       |
| 0.526         | 0.4184 | 200  | 0.5562          | -1.3989        | -2.5435          | 0.7617             | 1.1446          | -517.0156      | -402.5200    | 1.7427          | 0.9802        |
| 0.5202        | 0.5230 | 250  | 0.5419          | -1.1425        | -2.3279          | 0.7891             | 1.1854          | -495.4537      | -376.8783    | 1.4380          | 0.6489        |
| 0.5054        | 0.6276 | 300  | 0.5450          | -1.3981        | -2.6883          | 0.7773             | 1.2901          | -531.4894      | -402.4424    | 2.2560          | 1.4771        |
| 0.497         | 0.7322 | 350  | 0.5302          | -1.6005        | -2.8675          | 0.7734             | 1.2670          | -549.4120      | -422.6754    | 2.2259          | 1.3704        |
| 0.5076        | 0.8368 | 400  | 0.5348          | -1.6133        | -2.9625          | 0.7891             | 1.3492          | -558.9131      | -423.9595    | 2.2785          | 1.3332        |
| 0.5092        | 0.9414 | 450  | 0.5341          | -1.6701        | -3.0297          | 0.7852             | 1.3596          | -565.6297      | -429.6380    | 2.3444          | 1.3858        |


### Framework versions

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