app.py

import gradio as gr
import numpy as np
import os
from huggingface_hub import hf_hub_download
from camel_tools.data import CATALOGUE
from camel_tools.tagger.default import DefaultTagger
from camel_tools.disambig.bert import BERTUnfactoredDisambiguator

def predict_label(text):
   
    ip = text.split()
    ip_len = [len(ip)]
    
    span_scores = extract_spannet_scores(span_model,ip,ip_len)
    span_pooled_scores = pool_span_scores(span_scores, ip_len)

    pos_tags = tagger.tag(ip)
    msa_span_scores = extract_spannet_scores(msa_span_model,ip,ip_len,pos=pos_tags)
    msa_pooled_scores = pool_span_scores(msa_span_scores, ip_len)

    ensemble_span_scores = [score for scores in [span_scores, msa_span_scores] for score in scores]
    ensemble_pooled_scores = pool_span_scores(ensemble_span_scores, ip_len)
    
    ent_scores = extract_ent_scores(entity_model,ip,ensemble_pooled_scores)
    combined_sequences, ent_pred_tags = pool_ent_scores(ent_scores, ip_len)

    
    # ops = [[i,o] for i,o in zip(ip,combined_sequences[-1])]
    
    return combined_sequences


if __name__ == '__main__':
    
    space_key = os.environ.get('key')
    filenames = ['network.py', 'layers.py', 'utils.py',
                 'representation.py', 'predict.py', 'validate.py']
    
    for file in filenames:
        hf_hub_download('nehalelkaref/stagedNER',
                    filename=file,
                    local_dir='src',
                    token=space_key)
        
    CATALOGUE.download_package("all",
                           recursive=True,
                           force=True,
                           print_status=True)
    
    from src.predict import extract_spannet_scores,extract_ent_scores,pool_span_scores,pool_ent_scores
    from src.network import SpanNet, EntNet
    from src.validate import entities_from_token_classes


    diasmbig = BERTUnfactoredDisambiguator.pretrained('msa')
    tagger = DefaultTagger(diasmbig, 'pos')
    
    span_path = 'models/span.model'
    msa_span_path = 'new_models/msa.best.model'
    entity_path= 'models/entity.msa.model'
    
    span_model = SpanNet.load_model(span_path)
    msa_span_model = SpanNet.load_model(msa_span_path)
    entity_model = EntNet.load_model(entity_path)
    

    
    with gr.Blocks(theme='finlaymacklon/smooth_slate') as iface:
        example_input=gr.Textbox(label="Input Example", lines=3)
        prediction=gr.Text(label="Predicted Entities")
        
        
        gr.Interface(fn=predict_label, inputs=example_input,
                    outputs=prediction,theme="smooth_slate",
                    title="Flat Entity Classification for Levantine Arabic")
        gr.Examples(
      examples=["النشرة الإخبارية الصادرة عن الأونروا رقم 113 (1986/1/8).",
               "صورة لمدينة أريحا القديمة :تل السلطان",
               "صورة اطفال مخيم للاجئين الفلسطينيين في لبنان"],
      inputs= example_input)
        
    iface.launch(show_api=False)