app.py

import os
from threading import Thread, Event
from typing import Iterator

import gradio as gr

import torch
from transformers import AutoModelForCausalLM, GemmaTokenizerFast, TextIteratorStreamer
DESCRIPTION = """\
# Monlam LLM v2.0.1
"""
path="TenzinGayche/tpo_v1.0.0_ep1_dpo_ft"
MAX_MAX_NEW_TOKENS = 2048
DEFAULT_MAX_NEW_TOKENS = 1024
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))

# Load the model and tokenizer
tokenizer = GemmaTokenizerFast.from_pretrained(path)
model = AutoModelForCausalLM.from_pretrained(path, torch_dtype=torch.float16).to("cuda")

model.config.sliding_window = 4096
model.eval()

# Create a shared stop event
stop_event = Event()

def generate(
    message: str,
    chat_history: list[dict],
    max_new_tokens: int = 2048,
    temperature: float = 0.6,
    top_p: float = 0.9,
    top_k: int = 50,
    repetition_penalty: float = 1.2,
    do_sample: bool= False
) -> Iterator[str]:
    # Clear the stop event before starting a new generation
    stop_event.clear()


    # Append the user's message to the conversation history
    conversation = chat_history.copy()
    if not conversation:
        conversation.extend([
            {
                "role": "user",
                "content": "ཁྱེད་རང་སྨོན་ལམ་མི་བཟོས་རིག་ནུས་ཤིག་ཡིན་པ་དང་ཁྱེད་རང་མི་བཟོས་རིག་ནུས་(AI)ཤིག་ཡིན།"
            },
            {
                "role": "assistant",
                "content": "ལགས་སོ། ང་ཡིས་ཁྱེད་ཀྱི་བཀའ་བཞིན་སྒྲུབ་ཆོག"
            }
        ])
    conversation.append({"role": "user", "content": message})

    input_ids = tokenizer.apply_chat_template(conversation, add_generation_prompt=True, return_tensors="pt")
    if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
        input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
        gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
    input_ids = input_ids.to(model.device)

    # Create a streamer to get the generated response
    streamer = TextIteratorStreamer(tokenizer, timeout=20.0, skip_prompt=True, skip_special_tokens=True)
    generate_kwargs = dict(
        {"input_ids": input_ids},
        streamer=streamer,
        max_new_tokens=max_new_tokens,

    )
    
    # Run generation in a background thread
    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()

    outputs = []
    for text in streamer:
        if stop_event.is_set():
            break  # Stop if the stop button is pressed
        outputs.append(text)
        yield "".join(outputs)

    # After generation, append the assistant's response to the chat history
    assistant_response = "".join(outputs)
    chat_history.append({"role": "assistant", "content": assistant_response})


# Define a function to stop the generation
def stop_generation():
    stop_event.set()

# Create the chat interface with additional inputs and the stop button
with gr.Blocks(css="style.css", fill_height=True) as demo:
    gr.Markdown(DESCRIPTION)

    # Create the chat interface
    chat_interface = gr.ChatInterface(
        fn=generate,
        examples=[
            ["Hello there! How are you doing?"],
            ["Can you explain briefly to me what is the Python programming language?"],
            ["Explain the plot of Cinderella in a sentence."],
            ["How many hours does it take a man to eat a Helicopter?"],
            ["Write a 100-word article on 'Benefits of Open-Source in AI research'"],
        ],
        cache_examples=False,
        type="messages",
    )
    

if __name__ == "__main__":
    demo.queue(max_size=20).launch(share=True)