Shotor (Llama 3 8B Instruction Tuned on Farsi)

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.

Usage

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

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: amirmasoud.ahkol@gmail.com
• Sahar Mirzapour: saharmirzapoursahar@gmail.com