app.py

from transformers import AutoTokenizer, AutoModel
import pandas as pd
import numpy as np
import random 
import torch
import os 
import gradio as gr

# Model
auth_token = os.environ.get("TOKEN_FROM_SECRET")
checkpoint = 'srota/job-bert-mini'
model = AutoModel.from_pretrained(checkpoint, token=auth_token)
tokenizer = AutoTokenizer.from_pretrained(checkpoint, token=auth_token)

# Data
titles = pd.read_csv('inventory.csv', usecols=['title'])['title'].tolist()
descriptions = pd.read_csv('inventory.csv', usecols=['description'])['description'].tolist()
with open('inventory.npy', 'rb') as f:     
    embeddings = np.load(f)

# Inference
def inference(query, top_k=5):
    with torch.no_grad():  
        inputs = tokenizer([query], padding=True, truncation=True, max_length=512, return_tensors='pt')        
    query_embedding = model(**inputs)['last_hidden_state'][:,0,:].detach().numpy()
    cosines = np.dot(query_embedding, embeddings.T)[0]
    indexes = np.argsort(cosines)[-top_k:]
    return '\n\n'.join(['*' + t for i, t in enumerate(titles) if i in indexes])

# Gradio
examples = [['Data Scientist'], ['Warehouse Worker'], ['Gardener'], ['Part-Time Cleaner'], ['Math Teacher'], ['Registered Nurse'], ['Line Cook'],['Night Porter'],['Dietitian'],['Planned Surveyor'],['Driving Instructor'],['Senior It Engineer'],['Stores Person'],['Dental Hygienist'],['Event Manager'],['Welder'],['Underwriter'],['Frontend Developer'],['Paralegal'],['Copywriter'],['Community Nurse'],['Courier'],['Personal Trainer'],['Night Porter'],['Pharmacist'],['Carpenter']]

demo = gr.Interface(
    fn=inference,
    title='Job Search', 
    description='Simulate a semantic search for retrieving job titles that match the user query (the match is performed between the user query and 15K job descriptions)',
    inputs=gr.Textbox(lines=1, placeholder='', label="User keyword"),
    outputs=gr.Textbox(lines=10, label="Relevant jobs"),
    examples=random.sample(examples, 10)
)

demo.launch()