---
model-index:
- name: Reproduced-tulu2-dpo-13b
  results: []
datasets:
- HuggingFaceH4/ultrafeedback_binarized
language:
- en
base_model: allenai/tulu-2-13b
---

# Model Card for Reproduced Tulu2 DPO 13B

- This repository provides a reproduction version of Tulu2-DPO-13B finetuned upon [Tulu2-13B](https://huggingface.co/allenai/tulu-2-13b) and [Ultrafeedback](https://huggingface.co/datasets/HuggingFaceH4/ultrafeedback_binarized). 
- Therefore, we obey all licenses mentioned in Tulu2's work.
- Check our codes for more details: https://github.com/LuJunru/LLM_Finetune/tree/DPO. The codes are built with [TRL](https://github.com/huggingface/trl/tree/main).

## Performance

| Model | Size | Alignment | MT-Bench (score) | AlpacaEval (win rate %) |
|-------------|-----|----|---------------|--------------|
| **Tulu2-13b** | **13B** | **SFT** | **6.70** | **78.9** |
| **Tulu2-dpo-13b** | **13B** | **DPO** | **7.00** | **89.5** |
| **Reproduced-Tulu2-dpo-13b** | **13B** | **DPO** | **?** | **?** |

## Input Format

The model is trained to use the following format (note the newlines):
```
<|user|>
Your message here!
<|assistant|>
```

For best results, format all inputs in this manner. **Make sure to include a newline after `<|assistant|>`, this can affect generation quality quite a bit.**

## Training hyperparameters

The following hyperparameters were used during DPO training:
- learning_rate: 1e-6 * sqrt(Num of Nodes)
- total_train_batch_size: 128 * Num of Nodes
- optimizer: AdamW with default values
- lr_scheduler_type: linear
- lr_scheduler_warmup_ratio: 0.1
- Weight Decay: 0.05
- num_epochs: 3.0