README.md

---
license: apache-2.0
language:
- fa
- en
library_name: transformers
pipeline_tag: text-generation
datasets:
- myrkur/persian-alpaca-deep-clean
---

# Shotor (Llama 3 8B Instruction Tuned on Farsi)

<a href="https://ibb.co/PwCN3VF"><img src="https://i.ibb.co/0hJc8zm/shotor.png" alt="shotor" border="0"></a>


Shotor is a Persian language model built upon the llama 3 8B architecture, a multilingual Large Language Model (LLM). It has been fine-tuned using supervised learning techniques and the Dora method for efficient fine-tuning. The model has been specifically tailored and trained on Persian datasets, particularly leveraging the dataset provided by [persian-alpaca-deep-clean](https://huggingface.co/datasets/myrkur/persian-alpaca-deep-clean).

## Usage

Here's a sample Python code snippet demonstrating how to use Shotor for text generation:

```python
import transformers
import torch

# Load the Shotor model
model_id = "myrkur/shotor"
pipeline = transformers.pipeline(
    "text-generation",
    model=model_id,
    model_kwargs={"torch_dtype": torch.bfloat16},
    device_map="auto",
)

# Define user messages
messages = [
    {"role": "user", "content": "علم بهتر است یا ثروت؟"},
]

# Apply chat template and generate text
prompt = pipeline.tokenizer.apply_chat_template(
        messages, 
        tokenize=False, 
        add_generation_prompt=True
)

terminators = [
    pipeline.tokenizer.eos_token_id,
    pipeline.tokenizer.convert_tokens_to_ids("<|eot_id|>")
]

outputs = pipeline(
    prompt,
    max_new_tokens=512,
    eos_token_id=terminators,
    do_sample=True,
    temperature=0.5,
    top_p=0.9,
    repetition_penalty=1.1
)
print(outputs[0]["generated_text"][len(prompt):])
```

## Contributions

Contributions to Shotor are welcome! Whether it's enhancing the model's capabilities, improving its performance on specific tasks, or evaluating its performance, your contributions can help advance Persian natural language processing.

## Contact
For questions or further information, please contact:

- Amir Masoud Ahmadi: [[email protected]](mailto:[email protected])
- Sahar Mirzapour: [[email protected]](mailto:[email protected])