Update model.py

model.py

@@ -1,28 +1,10 @@
-import csv
-import io
-import os
-import requests
-import json
-import html  # For escaping HTML characters
-from bs4 import BeautifulSoup
-from openai import OpenAI 
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import torch
 
-# Initialize OpenAI API with Nvidia's Llama 3.1 70b nemotron model
-client = OpenAI(
-    base_url="https://integrate.api.nvidia.com/v1",
-    api_key=os.environ.get("KEY")
-)
-
-def clean_text_output(text):
+def modelFeedback_direct(ats_score, resume_data, job_description):
     """
-    Cleans the output to handle HTML characters and unwanted tags.
+    Generate ATS feedback by loading model and tokenizer directly.
     """
-    text = html.unescape(text)  # Unescape HTML entities
-    soup = BeautifulSoup(text, 'html.parser')  # Use BeautifulSoup to handle HTML tags
-    cleaned_text = soup.get_text(separator="\n").strip()  # Remove tags and handle newlines
-    return cleaned_text
-
-def modelFeedback(ats_score, resume_data, job_description):
     input_prompt = f"""
         You are now an ATS Score analyzer and given ATS Score is {int(ats_score * 100)}%. 
         Your task is to provide a comprehensive review and feedback based on the ATS score.
@@ -79,25 +61,25 @@ def modelFeedback(ats_score, resume_data, job_description):
         IMPORTANT: The output should be as normal organised text not in any other format (don't give markdown text) and focus more on resume matching part that general resume review.
         """
 
+    # Load tokenizer and model
+    tokenizer = AutoTokenizer.from_pretrained("nvidia/Llama-3.1-Nemotron-70B-Instruct-HF")
+    model = AutoModelForCausalLM.from_pretrained("nvidia/Llama-3.1-Nemotron-70B-Instruct-HF")
+
+    # Tokenize the input
+    inputs = tokenizer.encode(input_prompt, return_tensors="pt")
 
+    # Generate the response
     try:
-        # Generate response using the OpenAI API
-        response = client.chat.completions.create(
-            model="nvidia/llama-3.1-nemotron-70b-instruct",  # Using Llama 3.1 70b
-            messages=[
-                {"role": "user", "content": input_prompt}
-            ],
-            temperature=0.01,  # Lowering temperature for precise output
-            top_p=0.7,  # Prioritize high-probability tokens
-            max_tokens=1500,  # Allow longer content
+        outputs = model.generate(
+            inputs,
+            max_length=1500,
+            temperature=0.01,
+            top_p=0.7,
+            pad_token_id=tokenizer.eos_token_id
         )
-
-        # Extract and clean the response
-        feedback_text = response.choices[0].message.content.strip()  # Corrected line
-        cleaned_feedback = clean_text_output(feedback_text)
-
+        response_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
+        cleaned_feedback = clean_text_output(response_text)
         return cleaned_feedback
-
-    except requests.exceptions.RequestException as e:
-        print(f"API request failed: {str(e)}")
-        return "Error: Unable to generate feedback."
+    except Exception as e:
+        print(f"Model generation error: {e}")
+        return "Error: Unable to generate feedback."