app.py

import gradio as gr
import os
import spaces
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
from threading import Thread


# Set an environment variable
HF_TOKEN = os.environ.get("HF_TOKEN", None)

    
# Load the tokenizer and model
tokenizer = AutoTokenizer.from_pretrained("Telugu-LLM-Labs/Indic-gemma-7b-finetuned-sft-Navarasa-2.0")
model = AutoModelForCausalLM.from_pretrained("Telugu-LLM-Labs/Indic-gemma-7b-finetuned-sft-Navarasa-2.0", device_map="auto")


@spaces.GPU(duration=120)
def gemma(message: str, 
              history: list, 
              temperature: float, 
              max_new_tokens: int
             ) -> str:
    """
    Generate a streaming response using the Gemma model.
    Args:
        message (str): The input message.
        history (list): The conversation history used by ChatInterface.
        temperature (float): The temperature for generating the response.
        max_new_tokens (int): The maximum number of new tokens to generate.
    Returns:
        str: The generated response.
    """
    conversation = []
    for user, assistant in history:
        conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
    conversation.append({"role": "user", "content": message})

    input_prompt = """
### Instruction:
You are an AI assistant. Engage in a conversation with the user and provide helpful responses.

### Input:
{}

### Response:
"""

    input_text = input_prompt.format(message)

    inputs = tokenizer([input_text], return_tensors="pt").to(model.device)
    
    streamer = TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True)

    generate_kwargs = dict(
        **inputs,
        streamer=streamer,
        max_new_tokens=max_new_tokens,
        do_sample=True,
        temperature=temperature,
    )
    # This will enforce greedy generation (do_sample=False) when the temperature is passed 0, avoiding the crash.             
    if temperature == 0:
        generate_kwargs['do_sample'] = False
        
    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()

    outputs = []
    for text in streamer:
        outputs.append(text)
        yield "".join(outputs)
        

# Gradio block
chatbot=gr.Chatbot(placeholder="Prompt away in your local language",height=500)

with gr.Blocks(fill_height=True) as demo:
    
    gr.ChatInterface(
        fn=gemma,
        chatbot=chatbot,
        fill_height=True,
        additional_inputs_accordion=gr.Accordion(label="⚙️ Parameters", open=False, render=False),
        additional_inputs=[
            gr.Slider(minimum=0,
                      maximum=1, 
                      step=0.1,
                      value=0.95, 
                      label="Temperature", 
                      render=False),
            gr.Slider(minimum=128, 
                      maximum=4096,
                      step=1,
                      value=512, 
                      label="Max new tokens", 
                      render=False ),
            ],
        examples=[
            ["Tell me a story of a crow in Malayalam"],
            ["Explain to me what is AI in Hindi"],
            ["मुझे एक कौवे की कहानी बताओ"],
            ["ఒక కాకి కథ చెప్పండి"],
            ["मला कावळ्याची गोष्ट सांगा"],
            ["مجھے کوے کی کہانی سناؤ"],
            ["কাউৰীৰ কাহিনী এটা কওকচোন"],
            ["मलाई कागको कथा सुनाउनुहोस्"],
            ["مون کي ڪانءَ جي ڪهاڻي ٻڌاءِ"],
            ["ஒரு காகத்தின் கதையைச் சொல்லுங்கள்"],
            ["ಒಂದು ಕಾಗೆಯ ಕಥೆ ಹೇಳು"],
            ["ഒരു കാക്കയുടെ കഥ പറയൂ"],
            ["મને કાગડાની વાર્તા કહો"],
            ["ਮੈਨੂੰ ਇੱਕ ਕਾਂ ਦੀ ਕਹਾਣੀ ਸੁਣਾਓ"],
            ["একটা কাকের গল্প বল"],
            ["ମୋତେ କାଉର କାହାଣୀ କୁହ |"]
            ],
        cache_examples=False,
                     )
    
if __name__ == "__main__":
    demo.launch()