main.py

from flask import Flask, request, jsonify
from langchain_community.llms import LlamaCpp
from sentence_transformers import SentenceTransformer
from transformers import AutoTokenizer, AutoModel
from sklearn.metrics.pairwise import cosine_similarity
from flask import Flask, request, jsonify
import torch
import os

app = Flask(__name__)

n_gpu_layers = 0
n_batch = 1024


llm = LlamaCpp(
    model_path="Phi-3-mini-4k-instruct-q4.gguf",  # path to GGUF file
    temperature=0.1,
    n_gpu_layers=n_gpu_layers,
    n_batch=n_batch,
    verbose=True,
    n_ctx=4096
)
model0 = AutoModel.from_pretrained('sentence-transformers/paraphrase-TinyBERT-L6-v2')

model = SentenceTransformer('sentence-transformers/paraphrase-TinyBERT-L6-v2')

file_size = os.stat('Phi-3-mini-4k-instruct-q4.gguf')
print("model size ====> :", file_size.st_size, "bytes")


@app.route('/cv', methods=['POST'])
def get_skills():
    cv_body = request.json.get('cv_body')

    # Simple inference example
    output = llm(
        f"<|user|>\n{cv_body}<|end|>\n<|assistant|>Can you list the skills mentioned in the CV?<|end|>",
        max_tokens=256,  # Generate up to 256 tokens
        stop=["<|end|>"], 
        echo=True,  # Whether to echo the prompt
    )

    return jsonify({'skills': output})

@app.get('/')
def health():
    return jsonify({'status': 'Worked'})

# we will make here post request to compare between lists of skills one has employee just one text and the other has the of jobs has many texts
# the llm will say the most similar job to the cv
@app.route('/compare', methods=['POST'])
def compare():
    jobs_skills = request.json.get('jobs_skills')
    employee_skills = request.json.get('employee_skills')
    
    if not isinstance(jobs_skills, list) or not all(isinstance(skill, str) for skill in jobs_skills):
        raise ValueError("jobs_skills must be a list of strings")
    
    # Encode job and employee skills
    job_embeddings = [model.encode(skill) for skill in jobs_skills]
    employee_embeddings = model.encode(employee_skills)
    
    # Calculate cosine similarity
    similarities = []
    employee_embedding_tensor = torch.tensor(employee_embeddings).unsqueeze(0)
    for job_embedding in job_embeddings:
        job_embedding_tensor = torch.tensor(job_embedding).unsqueeze(0)
        similarity = cosine_similarity(employee_embedding_tensor, job_embedding_tensor)
        similarities.append(similarity.item())
    
    # Find the job with highest similarity
    max_similarity_index = similarities.index(max(similarities))
    max_similarity_job = jobs_skills[max_similarity_index]
    
    return jsonify({'job': max_similarity_job, 'similarity': max(similarities)})

if __name__ == '__main__':
    app.run()