Spaces:
Sleeping
Sleeping
| # imports | |
| import json | |
| import time | |
| import gradio as gr | |
| from transformers import AutoTokenizer, AutoModel | |
| import openai | |
| # pytorch library | |
| import torch | |
| import torch.nn.functional as f | |
| from fuzzywuzzy import process | |
| from roles_list import roles | |
| from openai import OpenAI | |
| # Load the model from the specified directory | |
| embed_store = {} | |
| model = 'sentence-transformers/all-MiniLM-L12-v2' | |
| sbert_model = AutoModel.from_pretrained(model) | |
| sbert_tokenizer = AutoTokenizer.from_pretrained(model) | |
| client = OpenAI( | |
| # defaults to os.environ.get("OPENAI_API_KEY") | |
| api_key="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", | |
| ) | |
| for role in roles: | |
| encoding = sbert_tokenizer(role, # the texts to be tokenized | |
| max_length=10, | |
| padding="max_length", | |
| return_tensors='pt' # return the tensors (not lists) | |
| ) | |
| with torch.no_grad(): | |
| # get the model embeddings | |
| embed = sbert_model(**encoding) | |
| embed = embed.pooler_output | |
| embed_store[role] = f.normalize(embed, p=2, dim=1) | |
| print("Model is ready for inference") | |
| def get_role_from_sbert(title): | |
| start_time = time.time() | |
| encoding = sbert_tokenizer(title, | |
| max_length=10, | |
| padding="max_length", | |
| return_tensors='pt' | |
| ) | |
| # Run the model prediction on the input data | |
| with torch.no_grad(): | |
| # get the model embeddings | |
| embed = sbert_model(**encoding) | |
| embed = embed.pooler_output | |
| store_cos = {} | |
| for role in embed_store: | |
| cos_sim = torch.nn.functional.cosine_similarity(f.normalize(embed, p=2, dim=1), embed_store[role]) | |
| store_cos[role] = round(cos_sim.item(), 3) | |
| # Get the top 3 items with the highest cosine similarity | |
| top_3_keys_values = sorted(store_cos.items(), key=lambda item: item[1], reverse=True) | |
| job_scores_str = '\n'.join([f"{job}: {score}" for job, score in top_3_keys_values]) | |
| end_time = time.time() | |
| execution_time = end_time - start_time | |
| # Convert to dictionary if needed or keep as list of tuples | |
| return job_scores_str + f" \nExecution time: {str(execution_time)}" | |
| def fuzzy_match(title): | |
| """ | |
| Find the best matches for a query from a list of choices using fuzzy matching. | |
| Parameters: | |
| - query: The search string. | |
| - choices: A list of strings to search through. | |
| - limit: The maximum number of matches to return. | |
| Returns: | |
| A list of tuples with the match and its score. Higher score means closer match. | |
| """ | |
| matches = process.extract(title, roles, limit=3) | |
| return matches | |
| def fuzzy_match_sbert(title): | |
| matches = fuzzy_match(title) | |
| sbert_results = get_role_from_sbert(title) | |
| new_list = [matches, sbert_results] | |
| return new_list | |
| demo = gr.Interface(fn=get_role_from_sbert, | |
| inputs=gr.Textbox(label="Job Title"), | |
| outputs=gr.Textbox(label="Role"), | |
| title="HackerRank Role Classifier") | |
| gr.close_all() | |
| demo.launch(server_name='0.0.0.0', server_port=8081, share=True) |