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

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: 0.5214
- Fcm Dpo/beta: 0.0836
- Fcm Dpo/q T: 0.3380
- Fcm Dpo/delta: -0.0050
- Fcm Dpo/margin: 11.8756
- Margin Dpo/margin Mean: 11.8756
- Margin Dpo/margin Std: 18.3875
- Logps/chosen: -96.2474
- Logps/rejected: -113.1114
- Logps/ref Chosen: -75.8693
- Logps/ref Rejected: -80.8577
- Logits/chosen: 0.3597
- Logits/rejected: 0.3046

## 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 | Fcm Dpo/beta | Fcm Dpo/q T | Fcm Dpo/delta | Fcm Dpo/margin | Margin Dpo/margin Mean | Margin Dpo/margin Std | Logps/chosen | Logps/rejected | Logps/ref Chosen | Logps/ref Rejected | Logits/chosen | Logits/rejected |
|:-------------:|:------:|:----:|:---------------:|:------------:|:-----------:|:-------------:|:--------------:|:----------------------:|:---------------------:|:------------:|:--------------:|:----------------:|:------------------:|:-------------:|:---------------:|
| 1.0502        | 0.3023 | 200  | 0.5717          | 0.0936       | 0.3525      | 0.0185        | 10.3675        | 10.3675                | 18.4204               | -87.9072     | -103.2631      | -75.8693         | -80.8577           | 0.4369        | 0.3875          |
| 1.0126        | 0.6047 | 400  | 0.5364          | 0.1033       | 0.3417      | 0.0098        | 9.4748         | 9.4748                 | 15.2862               | -92.9376     | -107.4008      | -75.8693         | -80.8577           | 0.3564        | 0.3008          |
| 1.0944        | 0.9070 | 600  | 0.5214          | 0.0836       | 0.3380      | -0.0050       | 11.8756        | 11.8756                | 18.3875               | -96.2474     | -113.1114      | -75.8693         | -80.8577           | 0.3597        | 0.3046          |


### Framework versions

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