from transformers import pipeline, TextIteratorStreamer
import torch
from threading import Thread
import gradio as gr
import spaces
import re

model_id = "openai/gpt-oss-20b"

pipe = pipeline(
    "text-generation",
    model=model_id,
    torch_dtype="auto",
    device_map="auto",
)

def format_conversation_history(chat_history):
    messages = []
    for item in chat_history:
        role = item["role"]
        content = item["content"]
        if isinstance(content, list):
            content = content[0]["text"] if content and "text" in content[0] else str(content)
        messages.append({"role": role, "content": content})
    return messages

@spaces.GPU(duration=60)
def generate_response(input_data, chat_history, max_new_tokens, system_prompt, temperature, top_p, top_k, repetition_penalty):
    new_message = {"role": "user", "content": input_data}
    system_message = [{"role": "system", "content": system_prompt}] if system_prompt else []
    processed_history = format_conversation_history(chat_history)
    messages = system_message + processed_history + [new_message]
    
    streamer = TextIteratorStreamer(pipe.tokenizer, skip_prompt=True, skip_special_tokens=True)
    generation_kwargs = {
        "max_new_tokens": max_new_tokens,
        "do_sample": True,
        "temperature": temperature,
        "top_p": top_p,
        "top_k": top_k,
        "repetition_penalty": repetition_penalty,
        "streamer": streamer
    }
    thread = Thread(target=pipe, args=(messages,), kwargs=generation_kwargs)
    thread.start()
    #streaming try #1
    buffer = ""
    full_response = ""
    for chunk in streamer:
        buffer += chunk
        parts = re.split(r'(\s+)', buffer)
        if re.match(r'\s+', parts[-1]) is not None:
            to_append = ''.join(parts)
            buffer = ""
        else:
            to_append = ''.join(parts[:-1])
            buffer = parts[-1]
        if to_append:
            full_response += to_append
            yield full_response
    if buffer:
        full_response += buffer
        yield full_response

demo = gr.ChatInterface(
    fn=generate_response,
    additional_inputs=[
        gr.Slider(label="Max new tokens", minimum=64, maximum=4096, step=1, value=2048),
        gr.Textbox(
            label="System Prompt",
            value="You are a helpful assistant. Reasoning: medium",
            lines=4,
            placeholder="Change system prompt"
        ),
        gr.Slider(label="Temperature", minimum=0.1, maximum=2.0, step=0.1, value=0.7),
        gr.Slider(label="Top-p", minimum=0.05, maximum=1.0, step=0.05, value=0.9),
        gr.Slider(label="Top-k", minimum=1, maximum=100, step=1, value=50),
        gr.Slider(label="Repetition Penalty", minimum=1.0, maximum=2.0, step=0.05, value=1.0)
    ],
    examples=[
        [{"text": "Explain Newton laws clearly and concisely"}],
        [{"text": "Write a Python function to calculate the Fibonacci sequence"}],
        [{"text": "What are the benefits of open weight AI models"}],
    ],
    cache_examples=False,
    type="messages",
    description="""
# gpt-oss-20b
Wait couple of seconds initially. You can adjust reasoning level in the system prompt like "Reasoning: high.
    """,
    fill_height=True,
    textbox=gr.Textbox(
        label="Query Input",
        placeholder="Type your prompt"
    ),
    stop_btn="Stop Generation",
    multimodal=False,
    theme=gr.themes.Soft()
)

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