ss

app.py

@@ -15,10 +15,11 @@ from docx import Document
 import csv
 from datasets import load_dataset
 import gc
-from huggingface_hub import InferenceClient
+from transformers import AutoModelForCausalLM, AutoTokenizer
 import time
 import faiss
 import re
+import openai
 
 # Download NLTK resources
 try:
@@ -46,7 +47,7 @@ with st.sidebar:
     
     # Advanced options
     st.subheader("Advanced Options")
-    top_k = st.number_input("Number of results to display", min_value=1, max_value=50, value=10, step=1)
+    top_k = st.selectbox("Number of results to display", options=[1, 5, 10, 20, 50], index=2)
     
     # LLM Settings
     st.subheader("LLM Settings")
@@ -80,12 +81,14 @@ if 'resume_texts' not in st.session_state:
     st.session_state.resume_texts = []
 if 'file_names' not in st.session_state:
     st.session_state.file_names = []
-if 'llm_client' not in st.session_state:
-    st.session_state.llm_client = None
 if 'explanations_generated' not in st.session_state:
     st.session_state.explanations_generated = False
 if 'current_job_description' not in st.session_state:
     st.session_state.current_job_description = ""
+if 'vllm_4b_endpoint' not in st.session_state:
+    st.session_state.vllm_4b_endpoint = "http://localhost:8001/v1"  # Qwen3-4B vLLM endpoint
+if 'vllm_14b_endpoint' not in st.session_state:
+    st.session_state.vllm_14b_endpoint = "http://localhost:8002/v1"  # Qwen3-14B vLLM endpoint
 
 @st.cache_resource
 def load_embedding_model():
@@ -112,28 +115,23 @@ def load_cross_encoder():
         st.error(f"❌ Error loading Cross-Encoder model: {str(e)}")
         return None
 
-def initialize_llm_client(hf_token=None):
-    """Initialize the LLM client for Qwen3-14B"""
-    try:
-        client = InferenceClient(
-            model="Qwen/Qwen3-14B",
-            token=hf_token if hf_token else None
-        )
-        return client
-    except Exception as e:
-        st.error(f"❌ Error initializing LLM client: {str(e)}")
-        return None
+def vllm_chat_completion(prompt, endpoint, max_tokens=200, temperature=0.7):
+    openai.api_base = endpoint
+    openai.api_key = "EMPTY"  # vLLM does not require a real key
+    response = openai.ChatCompletion.create(
+        model="Qwen/Qwen3-4B" if "4b" in endpoint else "Qwen/Qwen3-14B",
+        messages=[{"role": "user", "content": prompt}],
+        max_tokens=max_tokens,
+        temperature=temperature,
+        stream=False
+    )
+    return response.choices[0].message.content.strip()
 
 class ResumeScreener:
     def __init__(self):
         # Load models
         self.embedding_model = load_embedding_model()
         self.cross_encoder = load_cross_encoder()
-        self.llm_client = None
-    
-    def set_llm_client(self, client):
-        """Set the LLM client"""
-        self.llm_client = client
     
     def extract_text_from_file(self, file_path, file_type):
         """Extract text from various file types"""
@@ -363,10 +361,6 @@ class ResumeScreener:
     def add_intent_scores(self, resume_texts, job_description, top_20_with_bm25):
         """Stage 4: Add LLM intent analysis scores"""
         try:
-            if not self.llm_client:
-                st.warning("LLM client not available. Using default intent scores.")
-                return [(idx, cross_score, bm25_score, 0.1) for idx, cross_score, bm25_score in top_20_with_bm25]
-            
             results_with_intent = []
             progress_bar = st.progress(0)
             
@@ -402,12 +396,10 @@ Response format:
 Intent: [Yes/Maybe/No]
 Reason: [Brief justification]"""
             
-            response = self.llm_client.text_generation(
+            response = vllm_chat_completion(
                 prompt,
-                max_new_tokens=100,
-                temperature=0.3,
-                top_p=0.9,
-                do_sample=True
+                st.session_state.vllm_4b_endpoint,
+                max_tokens=100
             )
             
             # Parse response
@@ -516,7 +508,7 @@ Reason: [Brief justification]"""
     
     def generate_llm_explanation(self, resume_text, job_description, score, skills, max_retries=3):
         """Generate detailed explanation using Qwen3-14B"""
-        if not self.llm_client:
+        if not st.session_state.vllm_14b_endpoint:
             return self.generate_simple_explanation(score, score, score, skills)
         
         # Truncate texts to manage token limits
@@ -544,12 +536,10 @@ Write a professional, detailed 150-word analysis for THIS INDIVIDUAL CANDIDATE:"
         
         for attempt in range(max_retries):
             try:
-                response = self.llm_client.text_generation(
+                response = vllm_chat_completion(
                     prompt,
-                    max_new_tokens=200,
-                    temperature=0.7,
-                    top_p=0.9,
-                    do_sample=True
+                    st.session_state.vllm_14b_endpoint,
+                    max_tokens=200
                 )
                 
                 # Extract the response and ensure it's about 150 words
@@ -600,17 +590,6 @@ st.markdown("---")
 # Initialize screener
 screener = ResumeScreener()
 
-# Initialize LLM client if enabled
-if use_llm_explanations:
-    if 'hf_token' in locals() and hf_token:
-        if st.session_state.llm_client is None:
-            st.session_state.llm_client = initialize_llm_client(hf_token)
-    else:
-        if st.session_state.llm_client is None:
-            st.session_state.llm_client = initialize_llm_client()
-    
-    screener.set_llm_client(st.session_state.llm_client)
-
 # Job Description Input
 st.header("📝 Step 1: Enter Job Description")
 job_description = st.text_area(
@@ -864,7 +843,7 @@ with col2:
     show_explanation_button = (
         st.session_state.results and 
         use_llm_explanations and 
-        st.session_state.llm_client and
+        st.session_state.vllm_14b_endpoint and
         not st.session_state.explanations_generated
     )
     
@@ -907,8 +886,8 @@ with col2:
     elif st.session_state.results and not use_llm_explanations:
         st.info("💡 Enable 'Generate AI Explanations' in sidebar to use this feature")
     
-    elif st.session_state.results and not st.session_state.llm_client:
-        st.warning("⚠️ LLM client not available. Check your Hugging Face token.")
+    elif st.session_state.results and not st.session_state.vllm_14b_endpoint:
+        st.warning("⚠️ LLM model not available. Check your Hugging Face token.")
 
 # Display Results
 if st.session_state.results:

requirements.txt

@@ -1,5 +1,5 @@
 streamlit==1.31.0
-transformers==4.51.0
+transformers>=4.51.0
 torch==2.1.2
 pdfplumber==0.10.1
 PyPDF2==3.0.1
@@ -10,10 +10,12 @@ rank-bm25==0.2.2
 pandas==2.1.3
 numpy==1.24.3
 tqdm==4.66.1
-huggingface-hub>=0.30.0
+huggingface-hub==0.30.0
 bitsandbytes==0.44.1
 accelerate==0.27.2
 datasets==2.18.0
 sentence-transformers==2.7.0
 plotly==5.18.0
-einops
+einops
+vllm>=0.8.5
+openai>=1.0.0