import gradio as gr
from transformers import AutoTokenizer, AutoModel
import torch
import json

# Load the tokenizer
model_name = "TuringsSolutions/TechLegalV1"
tokenizer = AutoTokenizer.from_pretrained(model_name)

# Load adapter configuration manually
adapter_config_url = "https://huggingface.co/TuringsSolutions/TechLegalV1/resolve/main/adapter_config.json"
adapter_model_url = "https://huggingface.co/TuringsSolutions/TechLegalV1/resolve/main/adapter_model.safetensors"

# Download and load the adapter configuration
adapter_config = torch.hub.download_url_to_file(adapter_config_url, 'adapter_config.json')
with open(adapter_config, 'r') as f:
    adapter_config = json.load(f)

# Initialize the model with the adapter configuration
model = AutoModel.from_pretrained(model_name, trust_remote_code=True)

# Load adapter weights
model.load_adapter(adapter_model_url, config=adapter_config)

# Function to make predictions
def predict(text):
    inputs = tokenizer(text, return_tensors="pt")
    with torch.no_grad():
        outputs = model(**inputs)
    return outputs.last_hidden_state.mean(dim=1).squeeze().tolist()

# Create a Gradio interface
iface = gr.Interface(
    fn=predict,
    inputs=gr.inputs.Textbox(lines=2, placeholder="Enter text here..."),
    outputs="json",
    title="Tech Legal Model",
    description="A model for analyzing tech legal documents."
)

# Launch the interface
if __name__ == "__main__":
    iface.launch()