---
library_name: transformers
base_model: W-61/llama-3-8b-base-sft-hh-helpful-8xh200
tags:
- alignment-handbook
- epsilon-dpo
- generated_from_trainer
datasets:
- Anthropic/hh-rlhf
model-index:
- name: llama-3-8b-base-epsilon-dpo-hh-helpful-8xh200-20260410-233108
  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-epsilon-dpo-hh-helpful-8xh200-20260410-233108

This model is a fine-tuned version of [W-61/llama-3-8b-base-sft-hh-helpful-8xh200](https://huggingface.co/W-61/llama-3-8b-base-sft-hh-helpful-8xh200) on the Anthropic/hh-rlhf dataset.
It achieves the following results on the evaluation set:
- Loss: 0.6443
- Rewards/chosen: -0.4430
- Rewards/rejected: -0.5884
- Rewards/accuracies: 0.6402
- Rewards/margins: 0.1454
- Logps/chosen: -208.1799
- Logps/rejected: -243.5746
- Logps/ref Chosen: -87.8236
- Logps/ref Rejected: -82.8189
- Logits/chosen: -0.8325
- Logits/rejected: -0.7259
- Kl/p Epsilon Steps: 0.6150
- Kl/n Epsilon Steps: 0.3850

## 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: 16
- eval_batch_size: 16
- seed: 42
- distributed_type: multi-GPU
- num_devices: 8
- total_train_batch_size: 128
- total_eval_batch_size: 128
- 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 | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/chosen | Logps/rejected | Logps/ref Chosen | Logps/ref Rejected | Logits/chosen | Logits/rejected | Kl/p Epsilon Steps | Kl/n Epsilon Steps |
|:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:------------:|:--------------:|:----------------:|:------------------:|:-------------:|:---------------:|:------------------:|:------------------:|
| 0.6236        | 0.2941 | 100  | 0.6636          | -0.3033        | -0.3895          | 0.6124             | 0.0862          | -127.6472    | -134.3018      | -87.8236         | -82.8189           | -1.2261       | -1.1807         | 0.5760             | 0.4236             |
| 0.5934        | 0.5882 | 200  | 0.6429          | -0.5156        | -0.6820          | 0.6324             | 0.1663          | -182.9084    | -209.3034      | -87.8236         | -82.8189           | -0.9817       | -0.8951         | 0.6037             | 0.3958             |
| 0.6156        | 0.8824 | 300  | 0.6443          | -0.4430        | -0.5884          | 0.6402             | 0.1454          | -208.1799    | -243.5746      | -87.8236         | -82.8189           | -0.8325       | -0.7259         | 0.6150             | 0.3850             |


### Framework versions

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