from flask import Flask, request, jsonify
from langchain_community.llms import LlamaCpp
from sentence_transformers import SentenceTransformer
from transformers import AutoTokenizer, AutoModel

# cosine_similarity
import torch
from torch.nn.functional import cosine_similarity
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.invoke(
        f"\n{cv_body}\nCan you list the skills mentioned in the CV?",
        max_tokens=256,  # Generate up to 256 tokens
        stop=[""], 
        echo=True,  # Whether to echo the prompt
    )

    return jsonify({'skills': output})

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

# Endpoint to compare between employee skills and job skills
@app.route('/compare', methods=['POST'])
def compare():
    employee_skills = request.json.get('employee_skills')
    jobs_skills = request.json.get('jobs_skills')

    if not isinstance(employee_skills, list) or not all(isinstance(skill, str) for skill in employee_skills):
        raise ValueError("employee_skills must be a list of strings")

    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")

    job_embeddings = model.encode(jobs_skills)
    employee_embeddings = model.encode(employee_skills)

    similarity_scores = []
    employee_embeddings_tensor = torch.from_numpy(employee_embeddings).unsqueeze(0)

    for i, job_e in enumerate(job_embeddings):
        job_e_tensor = torch.from_numpy(job_e).unsqueeze(0)
        similarity_score = cosine_similarity(employee_embeddings_tensor, job_e_tensor, dim=1)
        similarity_scores.append({"job": jobs_skills[i], "similarity_score": similarity_score.tolist()[0]})

    return jsonify(similarity_scores)

# Endpoint to compare job posts with employee skills
@app.route('/compare_jop', methods=['POST'])
def compare_jop():
    employee_skills = request.json.get('post')
    jobs_skills = request.json.get('employee_skills')

    if not isinstance(employee_skills, list) or not all(isinstance(skill, str) for skill in employee_skills):
        raise ValueError("employee_skills must be a list of strings")

    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")

    job_embeddings = model.encode(jobs_skills)
    employee_embeddings = model.encode(employee_skills)

    similarity_scores = []
    employee_embeddings_tensor = torch.from_numpy(employee_embeddings).unsqueeze(0)

    for i, job_e in enumerate(job_embeddings):
        job_e_tensor = torch.from_numpy(job_e).unsqueeze(0)
        similarity_score = cosine_similarity(employee_embeddings_tensor, job_e_tensor, dim=1)
        similarity_scores.append({"job": jobs_skills[i], "similarity_score": similarity_score.tolist()[0]})

    return jsonify(similarity_scores)

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