from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
import spaces
import gradio as gr
from transformers import BitsAndBytesConfig
import os
from huggingface_hub import login

login(token=os.environ["HUGGINGFACE_TOKEN"])

model_name = "mistralai/Mistral-7B-v0.1"


tokenizer = AutoTokenizer.from_pretrained(model_name,use_auth_token=True)

quantization_config = BitsAndBytesConfig(
    load_in_4bit=True,  # Set `True` for 4-bit, `False` for 8-bit
    bnb_4bit_compute_dtype=torch.float16,
    bnb_4bit_use_double_quant=True
)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    quantization_config=quantization_config,
    device_map="auto"  # Automatically place model on GPU
)


# Inference Function
@spaces.GPU
def generate_text(prompt):
    inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
    output = model.generate(**inputs, max_new_tokens=100)
    return tokenizer.decode(output[0], skip_special_tokens=True)

# Example Usage

@spaces.GPU
def chat_with_llm(prompt):
    return generate_text(prompt)

gr.Interface(fn=chat_with_llm, inputs="text", outputs="text", title="Mistral Chatbot").launch()