README.md

---
language:
- en
- fa
---

<p align="center">
  <picture>
    <img alt="Hugging Face Transformers Library" src="https://i.postimg.cc/VN4F7WRC/Untitled-design-modified.png" width="1000" height="450" style="max-width: 100%;">
  </picture>
</p>

<h4 align="center">
    <p>
        <a href="https://huggingface.co/aidal/Persian-Mistral-7B#model-description">Model description</a> |
        <a href="https://huggingface.co/aidal/Persian-Mistral-7B#example-output">Example output</a> |
        <a href="https://huggingface.co/aidal/Persian-Mistral-7B#banchmark-results">Banchmark results</a> |
        <a href="https://huggingface.co/aidal/Persian-Mistral-7B#how-to-use">How to use</a> |
        <a href="https://huggingface.co/aidal/Persian-Mistral-7B#training-and-finetuning">Training and finetuning</a>
    </p>
</h4>

----

# Model description

>Persian-mistral is the fintuned version of mistral-7b that design for persian QA and nlp tasks
----

# Example output:

**Example 1:**
- Input: "درمان اصلی برای افراد مبتلا او آر اس (ORS)، جایگزینی مایعات و الکترولیت ها در بدن"
- Output: "درمان اصلی برای افراد مبتلا او آر اس (ORS)، جایگزینی مایعات و الکترولیت ها در بدن است. که به طور معمول از طریق تزریق وریدی استفاده می شود."

**Example 2:**
- Input: "سال ۱۹۴۴ متفقین به فرانسه اشغال شده توسط آلمان، در عملیاتی در نرماندی حمله کرده و"
- Output: "سال ۱۹۴۴ متفقین به فرانسه اشغال شده توسط آلمان، در عملیاتی در نرماندی حمله کرده و 150,000 نفر از آنها را کشتند."
----
# Banchmark results

| model         | dataset           | score  |
|---------------|-------------------|--------|
| base-model-7b | ARC-easy          |41.92%  |
| base-model-7b | ARC-easy          |39.12%  |
| fa-model-7b   | ARC-easy          |37.89%  |
| base-model-7b | ARC-challenge     |37.12%  |
| fa-model-7b   | ARC-challenge     |39.29%  |

----
# How to use

```python
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("aidal/Persian-Mistral-7B")
model = AutoModelForCausalLM.from_pretrained("aidal/Persian-Mistral-7B")
input_text = "پایتخت ایران کجاست؟"
input_ids = tokenizer(input_text, return_tensors="pt")
outputs = model.generate(**input_ids)
print(tokenizer.decode(outputs[0]))
```
----
# Training and finetuning
- **Extend tokenzer:** The base Mistral tokenizer does not support Persian. As an initial step, we trained a SentencePiece tokenizer on the Farsi Wikipedia corpus and subsequently integrated it with the Mistral tokenizer.
- **Pre-training:**  In the following step, we expanded the embedding layer of the base model to match the size of the Persian tokenizer. We then employed the LoRA method to train the model on three distinct datasets: Wikipedia-Farsi, an Islamic book collection, and content from Khamenei.ir.
<p align="center">
  <picture>
    <img alt="Hugging Face Transformers Library" src="https://i.postimg.cc/LXSD4HnZ/Stakehozlder-Map-1-page-0001-modified.png" width="270" height="270" style="max-width: 100%;">
  </picture>
</p>
<p align="center" style="font-size: 13px;">Wiki-farsi:183M tokens, Islamic books:55M tokens, Khamenei.ir:9M tokens</p>
- **Instruction Fine-tuning:** For the final step, we fine-tuned the model using the LoRA method on a translated version of the Stanford-alpaca to enhance the model's question-answering capabilities.
This diagram illustrates the steps described above:
<p align="center">
  <picture>
    <img alt="Hugging Face Transformers Library" src="https://i.postimg.cc/yY4dkwvT/Stakehozlder-Map-page-0001-modified.png" width="400" height="500" style="max-width: 100%;">
  </picture>
</p>