app.py

import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
import os

os.system('mkdir buff')

# Load model and tokenizer
model_name = "abhifdsdf/good_disease_8bit"  # Replace with your Hugging Face model repo name or local path
model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", low_cpu_mem_usage=True, offload_folder="buff")
tokenizer = AutoTokenizer.from_pretrained(model_name)

# Define inference function
def generate_text(input_text):
    inputs = tokenizer(input_text, return_tensors="pt")
    outputs = model.generate(inputs['input_ids'], max_length=50)
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

# Create Gradio interface
iface = gr.Interface(fn=generate_text, inputs="text", outputs="text", live=True)

# Launch the app
iface.launch()