app.py

import gradio as gr
from huggingface_hub import InferenceClient

"""
Copied from inference in colab notebook
"""

from transformers import LlamaForCausalLM, LlamaTokenizer
import torch

# Load model and tokenizer globally to avoid reloading for every request
model_path = "llama_lora_model_1"

# Load tokenizer
tokenizer = LlamaTokenizer.from_pretrained(model_path)

# Load model
model = LlamaForCausalLM.from_pretrained(
    model_path,
    torch_dtype=torch.float32,  # Adjust based on your environment
    device_map="cpu"  # Use CPU for inference
)

# Define the response function
def respond(
    message: str,
    history: list[tuple[str, str]],
    system_message: str,
    max_tokens: int,
    temperature: float,
    top_p: float,
):
    # Combine system message and history into a single prompt
    messages = [{"role": "system", "content": system_message}]
    for val in history:
        if val[0]:
            messages.append({"role": "user", "content": val[0]})
        if val[1]:
            messages.append({"role": "assistant", "content": val[1]})
    messages.append({"role": "user", "content": message})
    
    # Create a single text prompt from the messages
    prompt = ""
    for msg in messages:
        if msg["role"] == "system":
            prompt += f"[System]: {msg['content']}\n\n"
        elif msg["role"] == "user":
            prompt += f"[User]: {msg['content']}\n\n"
        elif msg["role"] == "assistant":
            prompt += f"[Assistant]: {msg['content']}\n\n"
    
    # Tokenize the prompt
    inputs = tokenizer(prompt, return_tensors="pt", truncation=True)
    input_ids = inputs.input_ids.to("cpu")  # Ensure input is on the CPU

    # Generate response
    output_ids = model.generate(
        input_ids,
        max_length=input_ids.shape[1] + max_tokens,
        temperature=temperature,
        top_p=top_p,
        do_sample=True,
    )
    
    # Decode the generated text
    generated_text = tokenizer.decode(output_ids[0], skip_special_tokens=True)

    # Extract the assistant's response from the generated text
    assistant_response = generated_text[len(prompt):].strip()

    # Yield responses incrementally (simulate streaming)
    response = ""
    for token in assistant_response.split():  # Split tokens by whitespace
        response += token + " "
        yield response.strip()


"""
For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
"""
# client = InferenceClient(model="https://huggingface.co/Heit39/llama_lora_model_1")



# def respond(
#     message,
#     history: list[tuple[str, str]],
#     system_message,
#     max_tokens,
#     temperature,
#     top_p,
# ):
#     messages = [{"role": "system", "content": system_message}]

#     for val in history:
#         if val[0]:
#             messages.append({"role": "user", "content": val[0]})
#         if val[1]:
#             messages.append({"role": "assistant", "content": val[1]})

#     messages.append({"role": "user", "content": message})

#     response = ""

#     for message in client.chat_completion(
#         messages,
#         max_tokens=max_tokens,
#         stream=True,
#         temperature=temperature,
#         top_p=top_p,
#     ):
#         token = message.choices[0].delta.content

#         response += token
#         yield response


"""
For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
"""
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95,
            step=0.05,
            label="Top-p (nucleus sampling)",
        ),
    ],
)


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