import gradio as gr
import spaces
from transformers import pipeline, AutoTokenizer, AutoModelForTokenClassification
from typing import List, Dict, Any
import torch

# Define the model and tokenizer
model_name = "kazalbrur/BanglaLegalNER"  # Ensure this model is suitable or update accordingly
tokenizer_name = "csebuetnlp/banglat5_banglaparaphrase"

# Load the tokenizer and model
tokenizer = AutoTokenizer.from_pretrained(tokenizer_name, use_fast=False)
model = AutoModelForTokenClassification.from_pretrained(model_name)

def merge_tokens(tokens: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
    merged_tokens = []
    for token in tokens:
        if merged_tokens and token['entity'].startswith('I-') and merged_tokens[-1]['entity'].endswith(token['entity'][2:]):
            last_token = merged_tokens[-1]
            last_token['word'] += token['word'].replace('##', '')
            last_token['end'] = token['end']
            last_token['score'] = (last_token['score'] + token['score']) / 2
        else:
            merged_tokens.append(token)
    return merged_tokens

# Determine device
device = 0 if torch.cuda.is_available() else -1

# Initialize Pipeline with the new model and tokenizer
get_completion = pipeline("ner", model=model, tokenizer=tokenizer, device=device)

@spaces.GPU(duration=120)
def ner(input: str) -> Dict[str, Any]:
    try:
        output = get_completion(input)
        merged_tokens = merge_tokens(output)
        return {"text": input, "entities": merged_tokens}
    except Exception as e:
        return {"text": input, "entities": [], "error": str(e)}

####### GRADIO APP #######
title = """<h1 id="title"> Bangla Legal Entity Recognition </h1>"""

description = """
- The model used for Recognizing entities [Bangla Legal NER](https://huggingface.co/kazalbrur/BanglaLegalNER).
"""

css = '''
h1#title {
  text-align: center;
}
'''

theme = gr.themes.Soft()
demo = gr.Blocks(css=css, theme=theme)

with demo:
    gr.Markdown(title)
    gr.Markdown(description)
    gr.Interface(
        fn=ner,
        inputs=[gr.Textbox(label="Enter Your Text to Find the Legal Entities", lines=20)],
        outputs=[gr.HighlightedText(label="Text with entities")],
        allow_flagging="never"
    )

demo.launch()