README.md

---
license: apache-2.0
datasets:
- argilla/distilabel-intel-orca-dpo-pairs
base_model: sethuiyer/Chikuma_10.7B
library_name: transformers
pipeline_tag: text-generation
tags:
  - dpo
  
---

# Chikuma_10.7B - V2 (Enhanced with DPO)

<p align="center">
  <img src="https://huggingface.co/sethuiyer/distilabled_Chikuma_10.7B/resolve/main/chikuma_v2.webp" height="256px" alt="Chikuma">
</p>


This model is the **DPO fine tuned version** of [Chikuma_10.7B](https://huggingface.co/sethuiyer/Chikuma_10.7B), which was a depth upscaled merge of:
* [sethuiyer/SynthIQ-7b](https://huggingface.co/sethuiyer/SynthIQ-7b)
* [openchat/openchat-3.5-0106](https://huggingface.co/openchat/openchat-3.5-0106)

The name "Chikuma" is inspired by the [Chikuma River](https://en.wikipedia.org/wiki/Shinano_River), the longest in Japan, known for its continuous flow and meandering path. 
This metaphorically represents the model's depth, fluidity, and adaptability in processing and understanding language.


# Dataset used for Fine Tuning
Dataset: `/argilla/distilabel-intel-orca-dpo-pairs`

The dataset was roughly ~3000 samples but they were high quality (according to the chosen_score).  

The following filters were applied to the original dataset:
```python
dataset = dataset.filter(
    lambda r:
        r["status"] != "tie" and
        r["chosen_score"] >= 8 and
        not r["in_gsm8k_train"]
)
```

# Chat Template
The chat template for Chikuma_10.7B - V2 is a modified version of ChatML, optimized for improved interaction and engagement:

```
<|im_start|>GPT4 Correct system:
{system} Always use <|end_of_turn|> when you want to end the answer. <|im_end|>
<|im_start|>GPT4 Correct user:
{user}<|im_end|>
<|im_start|>GPT4 Correct Assistant:
{asistant}<|im_end|>
```

## Nous Benchmark Evaluation
| Model                         | AGIEval | GPT4All | TruthfulQA | Bigbench | Average |
|-------------------------------|---------|---------|------------|----------|---------|
| SynthIQ-7b                    | 42.67   | 73.71   | 56.51      | 44.59    | 54.37   |
| openchat/openchat-3.5-0106    | 44.17   | 73.72   | 52.53      | 44.4     | 53.71   |
| Chikuma_10.7B                 | 42.41   | 73.41   | 56.69      | 43.5     | 54.00   |
| **distilabled_Chikuma_10.7B** | **42.77** | **73.81** | **58.83**  | **44.83** | **55.06** |

# OpenLLM Leaderboard

| Benchmark Name | Performance |
|----------------|-------------|
| ARC            | 66.38       |
| HellaSwag      | 85          |
| MMLU           | 65.27       |
| TruthfulQA     | 58.83       |
| Winogrande     | 78.77       |
| GSM8K          | 63.68       |
| **Average**    | **69.65**   |


### Training Environment
- Hardware: Single A100 80GB GPU in a runpod, utilized for approximately 1.5 hours.
- Training Script: Accessible via [Google Colab Notebook](https://colab.research.google.com/drive/15iFBr1xWgztXvhrj5I9fBv20c7CFOPBE?usp=sharing). Special thanks to [mlabonne](https://huggingface.co/mlabonne) for providing the template.


## Usage

```python
# Format prompt
from transformers import AutoModelForCausalLM, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained(new_model)

# Create pipeline
pipeline = transformers.pipeline(
    "text-generation",
    model=new_model,
    tokenizer=tokenizer,
    device="cuda"
)

# Generate text

message = [
    {"role": "system", "content": "You are a helpful assistant chatbot."},
    {"role": "user", "content": "Who invented LLMs?"}
]

prompt = tokenizer.apply_chat_template(message, add_generation_prompt=True, tokenize=False)

sequences = pipeline(
    prompt,
    max_new_tokens=512
)
print(sequences[0]['generated_text'])
```

## Acknowledgements

A heartfelt appreciation goes to the vibrant open-source community, particularly:

* The Intel team for publishing a great open dataset and show how well it worked in the first place 
* Teknium and NousResearch for their awesome work and models.
* Maxime for sharing such great resources.
* Argilla for publishing argilla/distilabel-intel-orca-dpo-pairs