updated gemma endpoint

config.py

@@ -1,41 +1,50 @@
+# === Imports ===
 import os
+import time
+import requests
 from dotenv import load_dotenv
 from supabase import create_client
-import requests
-import time
-from sentence_transformers import SentenceTransformer  # Import the transformer model
+from sentence_transformers import SentenceTransformer
+from openai import OpenAI
 
-# Load environment variables from .env file
+# === Load Environment Variables ===
 load_dotenv()
 
-# Supabase API Config
+# === Supabase Configuration ===
 SUPABASE_URL = "https://lmpazoxzucnlqqxjoihi.supabase.co"
 SUPABASE_KEY = os.getenv("SUPABASE_API_KEY")
 if not SUPABASE_KEY:
     raise ValueError("SUPABASE_KEY is not set in the environment variables.")
 supabase = create_client(SUPABASE_URL, SUPABASE_KEY)
 
-# Load Sentence Transformer Model (scoring)
+# === Embedding Model for Scoring ===
 embedding_model = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
 
-HF_MODELS = {
-    "pegasus": "https://router.huggingface.co/hf-inference/models/google/pegasus-xsum",
-    "gemma": "https://router.huggingface.co/nebius/v1/chat/completions"  
-
-}
-
+# === Hugging Face API Configuration ===
 HF_API_TOKEN = os.getenv("HF_API_TOKEN")
 if not HF_API_TOKEN:
     raise ValueError("Missing Hugging Face API key. Check your .env file.")
-
 HF_HEADERS = {"Authorization": f"Bearer {HF_API_TOKEN}"}
 
+# === Hugging Face Model Endpoints ===
+HF_MODELS = {
+    "pegasus": "https://router.huggingface.co/hf-inference/models/google/pegasus-xsum",
+    "gemma": "tgi"  # Used as the model name with OpenAI-compatible client
+}
+
+# === OpenAI-Compatible Client (for Gemma) ===
+client = OpenAI(
+    base_url="https://vzwjawyxvu030jsw.us-east-1.aws.endpoints.huggingface.cloud/v1/",
+    api_key=HF_API_TOKEN,
+)
+
+# === Optional: General Query Helper (for non-chat models like pegasus) ===
 def query(payload, model="pegasus", retries=5, delay=5):
     """
-    Sends a request to the Hugging Face API with retries and better error handling.
+    Sends a request to the Hugging Face API with retries and error handling.
     """
     if model not in HF_MODELS:
-        raise ValueError(f"Invalid model name: {model}. Choose 'pegasus' for summarization.")
+        raise ValueError(f"Invalid model name: {model}. Available: {list(HF_MODELS.keys())}")
 
     api_url = HF_MODELS[model]
 
@@ -44,15 +53,13 @@ def query(payload, model="pegasus", retries=5, delay=5):
             response = requests.post(api_url, headers=HF_HEADERS, json=payload, timeout=10)
 
             if response.status_code == 401:
-                print(f"❌ API Key issue: Check HF_API_TOKEN. Unauthorized (401).")
+                print("❌ Unauthorized (401). Check HF_API_TOKEN.")
                 return None
-
             if response.status_code == 402:
-                print(f"💰 Payment Required (402). Free tier may not support this model.")
+                print("💰 Payment Required (402). Free tier may not support this model.")
                 return None
-
             if response.status_code in [500, 503]:
-                print(f"⚠️ Server error ({response.status_code}) on attempt {attempt + 1}. Retrying in {delay} seconds...")
+                print(f"⚠️ Server error ({response.status_code}) on attempt {attempt + 1}. Retrying in {delay}s...")
                 time.sleep(delay)
                 continue
 
@@ -60,11 +67,10 @@ def query(payload, model="pegasus", retries=5, delay=5):
             return response.json()
 
         except requests.exceptions.Timeout:
-            print(f"⏳ Timeout error on attempt {attempt + 1}. Retrying...")
+            print(f"⏳ Timeout on attempt {attempt + 1}. Retrying in {delay}s...")
             time.sleep(delay)
-
         except requests.exceptions.RequestException as e:
-            print(f"❌ API Request Failed: {e}")
+            print(f"❌ Request failed: {e}")
             time.sleep(delay)
 
     print("🚨 All retry attempts failed.")

utils.py

@@ -18,20 +18,21 @@ from fuzzywuzzy import fuzz
 from sentence_transformers import SentenceTransformer, util
 from sklearn.feature_extraction.text import TfidfVectorizer
 from huggingface_hub import InferenceClient
+from openai import OpenAI
 
 # Local Configuration
 from config import (
     SUPABASE_URL, SUPABASE_KEY, HF_API_TOKEN, HF_HEADERS,
-    supabase, HF_MODELS, query, embedding_model
+    supabase, HF_MODELS, query, embedding_model, client
 )
 
 # === Initialization ===
 
-# Hugging Face inference client for Gemma model
-client = InferenceClient(
-    model="google/gemma-1.1-7b-it",
-    token=HF_API_TOKEN
-)
+# # Hugging Face inference client for Gemma model
+# client = InferenceClient(
+#     model="tgi",
+#     token=HF_API_TOKEN
+# )
 
 # Load or download spaCy model
 try:
@@ -173,18 +174,26 @@ def summarize_resume(resume_text):
     )
 
     try:
-        response = client.chat_completion(
+        response = client.chat.completions.create(
+            model="tgi",
             messages=[{"role": "user", "content": prompt}],
             temperature=0.5,
             max_tokens=300,
         )
         result = response.choices[0].message.content.strip()
-        return result
+
+        # Clean up generic lead-ins from the model
+        cleaned = re.sub(
+            r"^(Sure,|Certainly,)?\s*(here is|here’s|this is)?\s*(the)?\s*(extracted)?\s*(professional)?\s*summary.*?:\s*",
+            "", result, flags=re.IGNORECASE
+        ).strip()
+
+        return cleaned
+
     except Exception as e:
         print(f"❌ Error generating structured summary: {e}")
         return "Summary unavailable due to API issues."
 
-
 # === Data Storage & Reporting ===
 
 def store_in_supabase(resume_text, score, candidate_name, email, summary):
@@ -242,11 +251,13 @@ def generate_interview_questions_from_summaries(candidates):
     )
 
     try:
-        response = client.chat_completion(
+        response = client.chat.completions.create(
+            model="tgi",
             messages=[{"role": "user", "content": prompt}],
             temperature=0.7,
-            max_tokens=500
-        )
+            max_tokens=500,
+)
+
         result = response.choices[0].message.content
 
         # Clean and normalize questions
@@ -260,14 +271,17 @@ def generate_interview_questions_from_summaries(candidates):
             if not q or re.match(r"^#+\s*", q):
                 continue
 
-            # Remove leading "Q1.", "1)", etc.
-            q = re.sub(r"^(?:Q?\d+[\.\)\-]?\s*)+", "", q)
+            # Remove leading bullets like "1.", "1)", "- 1.", etc.
+            q = re.sub(r"^(?:[-*]?\s*)?(?:Q?\d+[\.\)\-]?\s*)+", "", q)
 
             # Remove markdown bold/italics (**, *, etc.)
             q = re.sub(r"[*_]+", "", q)
+            
+            # Remove duplicate trailing punctuation
+            q = q.strip(" .")
 
             questions.append(q.strip())
-
+            
         return [f"Q{i+1}. {q}" for i, q in enumerate(questions[:5])] or ["⚠️ No questions generated."]
     
     except Exception as e: