Update main.py

main.py

@@ -1,74 +1,73 @@
 from flask import Flask, request, jsonify
 from langchain_community.llms import LlamaCpp
 from sentence_transformers import SentenceTransformer
-from transformers import AutoTokenizer, AutoModel
-
-# cosine_similarity
+from transformers import AutoModel
 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
+    model_path="Phi-3-mini-4k-instruct-q4.gguf",  
     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')
 
+# Initializing sentence transformer model
 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")
+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
+        f"\n{cv_body}\nCan you list the skills mentioned in the CV?",
+        max_tokens=256,  
+        stop=[""], 
+        echo=True,  
     )
 
     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():
-    employee_skills = request.json.get('employee_skills')  # CV text
-    jobs_skills = request.json.get('jobs_skills')  # List of job skills
-    
+    employee_skills = request.json.get('employee_skills')
+    jobs_skills = request.json.get('jobs_skills')
+
+    # Validation
     if not isinstance(jobs_skills, list) or not all(isinstance(skill, str) for skill in jobs_skills):
-        raise ValueError("The jobs_skills must be a list of strings")
-    
-    # Convert texts to embeddings arrays
-    employee_embedding = np.array([model.encode(employee_skills)])
-    job_embeddings = np.array([model.encode(skill) for skill in jobs_skills])
+        raise ValueError("jobs_skills must be a list of strings")
 
-    # Calculate similarity using cosine similarity
-    similarities = cosine_similarity(employee_embedding, job_embeddings)[0]
+    # Encoding skills into embeddings
+    job_embeddings = model.encode(jobs_skills)
+    employee_embeddings = model.encode(employee_skills)
 
-    # Find the most similar job and its corresponding similarity score
-    max_similarity = np.max(similarities)
-    most_similar_index = np.argmax(similarities)
-    most_similar_job = jobs_skills[most_similar_index]
+    # Computing cosine similarity between employee skills and each job
+    similarity_scores = []
+    employee_embeddings_tensor = torch.from_numpy(employee_embeddings).unsqueeze(0)
 
-    return jsonify({'job': most_similar_job, 'similarity_score': max_similarity})
+    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.item()})
 
+    return jsonify(similarity_scores)
 if __name__ == '__main__':
     app.run()