---
library_name: transformers
base_model: W-61/llama-3-8b-base-sft-hh-harmless-4xh200
tags:
- alignment-handbook
- margin-dpo
- generated_from_trainer
datasets:
- Anthropic/hh-rlhf
model-index:
- name: llama-3-8b-base-margin-dpo-hh-harmless
  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. -->

# llama-3-8b-base-margin-dpo-hh-harmless

This model is a fine-tuned version of [W-61/llama-3-8b-base-sft-hh-harmless-4xh200](https://huggingface.co/W-61/llama-3-8b-base-sft-hh-harmless-4xh200) on the Anthropic/hh-rlhf dataset.
It achieves the following results on the evaluation set:
- Loss: 0.5259
- Margin Dpo/margin Mean: 9.3649
- Margin Dpo/margin Std: 14.8097
- Logps/chosen: -92.0386
- Logps/rejected: -106.0930
- Logps/ref Chosen: -74.8595
- Logps/ref Rejected: -79.5490
- Logits/chosen: 0.3798
- Logits/rejected: 0.3285

## 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: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 1

### Training results

| Training Loss | Epoch  | Step | Validation Loss | Margin Dpo/margin Mean | Margin Dpo/margin Std | Logps/chosen | Logps/rejected | Logps/ref Chosen | Logps/ref Rejected | Logits/chosen | Logits/rejected |
|:-------------:|:------:|:----:|:---------------:|:----------------------:|:---------------------:|:------------:|:--------------:|:----------------:|:------------------:|:-------------:|:---------------:|
| 1.3342        | 0.1512 | 100  | 0.6557          | 1.4205                 | 4.9786                | -79.7014     | -85.8115       | -74.8595         | -79.5490           | 0.2556        | 0.2183          |
| 0.9165        | 0.3023 | 200  | 0.5447          | 7.4721                 | 12.5600               | -86.5507     | -98.7123       | -74.8595         | -79.5490           | 0.3345        | 0.2868          |
| 0.9692        | 0.4535 | 300  | 0.5345          | 9.3794                 | 14.9738               | -93.1794     | -107.2484      | -74.8595         | -79.5490           | 0.4017        | 0.3507          |
| 1.084         | 0.6047 | 400  | 0.5337          | 8.8635                 | 14.3566               | -91.2627     | -104.8157      | -74.8595         | -79.5490           | 0.3912        | 0.3394          |
| 1.0037        | 0.7559 | 500  | 0.5277          | 9.5078                 | 15.0672               | -92.1725     | -106.3698      | -74.8595         | -79.5490           | 0.3937        | 0.3419          |
| 1.0459        | 0.9070 | 600  | 0.5259          | 9.3649                 | 14.8097               | -92.0386     | -106.0930      | -74.8595         | -79.5490           | 0.3798        | 0.3285          |


### Framework versions

- Transformers 4.51.0
- Pytorch 2.3.1+cu121
- Datasets 2.21.0
- Tokenizers 0.21.4