app.py

import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
import safetensors.torch

# Load base model and tokenizer
base_model_name = "AIDC-AI/Marco-o1"
tokenizer = AutoTokenizer.from_pretrained(base_model_name)
model = AutoModelForCausalLM.from_pretrained(base_model_name)

# Load your lightweight safetensor
safetensor_path = "adapter_model.safetensors"
model.load_state_dict(safetensors.torch.load_file(safetensor_path), strict=False)

def predict(text):
    inputs = tokenizer.encode(text, return_tensors="pt")
    outputs = model.generate(inputs)
    prediction = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return prediction

demo = gr.Interface(fn=predict, inputs="text", outputs="text")
demo.launch()