Update app.py

app.py

@@ -1,19 +1,25 @@
-import gradio as gr
+import os
 import faiss
 import numpy as np
-import os
 from sentence_transformers import SentenceTransformer
-from huggingface_hub import InferenceClient
+from huggingface_hub import HfApi, hf_hub_download, login, whoami
 
-# Load embedding model
-model = SentenceTransformer("all-MiniLM-L6-v2")
+# 🔹 Hugging Face Repository Details
+HF_REPO_ID = "tstone87/repo"  # Your repo
+HF_TOKEN = os.getenv("HF_TOKEN")  # Retrieve token securely from environment variable
 
-# File paths
-TEXT_FILE = "combined_text_documents.txt"
+if not HF_TOKEN:
+    raise ValueError("❌ ERROR: Hugging Face token not found. Add it as a secret in the Hugging Face Space settings.")
+
+# 🔹 Authenticate with Hugging Face
+login(token=HF_TOKEN)
+
+# 🔹 File Paths
 EMBEDDINGS_FILE = "policy_embeddings.npy"
 INDEX_FILE = "faiss_index.bin"
+TEXT_FILE = "combined_text_documents.txt"
 
-# Load policy text from file
+# 🔹 Load policy text from file
 if os.path.exists(TEXT_FILE):
     with open(TEXT_FILE, "r", encoding="utf-8") as f:
         POLICY_TEXT = f.read()
@@ -22,35 +28,88 @@ else:
     print("❌ ERROR: combined_text_documents.txt not found! Ensure it's uploaded.")
     POLICY_TEXT = ""
 
-# Split text into chunks for FAISS indexing
+# 🔹 Sentence Embedding Model (Optimized for Speed)
+model = SentenceTransformer("all-MiniLM-L6-v2")
+
+# 🔹 Split policy text into chunks for FAISS indexing
 chunk_size = 500
 chunks = [POLICY_TEXT[i:i+chunk_size] for i in range(0, len(POLICY_TEXT), chunk_size)] if POLICY_TEXT else []
 
-# Check if precomputed embeddings and FAISS index exist
+# 🔹 Function to Upload FAISS Files to Hugging Face Hub
+def upload_faiss_to_hf():
+    api = HfApi()
+
+    if os.path.exists(EMBEDDINGS_FILE):
+        print("📤 Uploading FAISS embeddings to Hugging Face...")
+        api.upload_file(
+            path_or_fileobj=EMBEDDINGS_FILE,
+            path_in_repo=EMBEDDINGS_FILE,
+            repo_id=HF_REPO_ID,
+            repo_type="dataset",
+            token=HF_TOKEN,
+        )
+
+    if os.path.exists(INDEX_FILE):
+        print("📤 Uploading FAISS index to Hugging Face...")
+        api.upload_file(
+            path_or_fileobj=INDEX_FILE,
+            path_in_repo=INDEX_FILE,
+            repo_id=HF_REPO_ID,
+            repo_type="dataset",
+            token=HF_TOKEN,
+        )
+
+    print("✅ FAISS files successfully uploaded to Hugging Face.")
+
+# 🔹 Function to Download FAISS Files from Hugging Face Hub if Missing
+def download_faiss_from_hf():
+    if not os.path.exists(EMBEDDINGS_FILE):
+        print("📥 Downloading FAISS embeddings from Hugging Face...")
+        hf_hub_download(repo_id=HF_REPO_ID, filename=EMBEDDINGS_FILE, local_dir=".", token=HF_TOKEN)
+
+    if not os.path.exists(INDEX_FILE):
+        print("📥 Downloading FAISS index from Hugging Face...")
+        hf_hub_download(repo_id=HF_REPO_ID, filename=INDEX_FILE, local_dir=".", token=HF_TOKEN)
+
+    print("✅ FAISS files downloaded from Hugging Face.")
+
+# 🔹 Check if FAISS Files Exist, Otherwise Download
 if os.path.exists(EMBEDDINGS_FILE) and os.path.exists(INDEX_FILE):
-    print("✅ Loading precomputed FAISS index and embeddings...")
+    print("✅ FAISS files found locally. Loading from disk...")
     embeddings = np.load(EMBEDDINGS_FILE)
     index = faiss.read_index(INDEX_FILE)
 else:
-    print("🚀 Generating embeddings and FAISS index (First-time setup)...")
-    if chunks:
-        embeddings = np.array([model.encode(chunk) for chunk in chunks])
-        np.save(EMBEDDINGS_FILE, embeddings)  # Save for future runs
-
-        # Use FAISS optimized index for faster lookup
-        d = embeddings.shape[1]
-        nlist = 10  # Number of clusters
-        index = faiss.IndexIVFFlat(faiss.IndexFlatL2(d), d, nlist)
-        index.train(embeddings)
-        index.add(embeddings)
-        index.nprobe = 2  # Speed optimization
-        faiss.write_index(index, INDEX_FILE)  # Save FAISS index
-        print("✅ FAISS index created and saved.")
-    else:
-        print("❌ ERROR: No text to index. Check combined_text_documents.txt.")
-        index = None
+    print("🚀 FAISS files not found! Downloading from Hugging Face...")
+    download_faiss_from_hf()
 
-# 🔹 Function to search FAISS
+    if os.path.exists(EMBEDDINGS_FILE) and os.path.exists(INDEX_FILE):
+        embeddings = np.load(EMBEDDINGS_FILE)
+        index = faiss.read_index(INDEX_FILE)
+    else:
+        print("🚀 No FAISS files found. Recomputing...")
+        if chunks:
+            embeddings = np.array([model.encode(chunk) for chunk in chunks])
+
+            # Save embeddings for future use
+            np.save(EMBEDDINGS_FILE, embeddings)
+
+            # Use FAISS optimized index for faster lookup
+            d = embeddings.shape[1]
+            nlist = 10  # Number of clusters
+            index = faiss.IndexIVFFlat(faiss.IndexFlatL2(d), d, nlist)
+            index.train(embeddings)
+            index.add(embeddings)
+            index.nprobe = 2  # Speed optimization
+
+            # Save FAISS index
+            faiss.write_index(index, INDEX_FILE)
+            upload_faiss_to_hf()  # Upload FAISS files to Hugging Face
+            print("✅ FAISS index created and saved.")
+        else:
+            print("❌ ERROR: No text to index. Check combined_text_documents.txt.")
+            index = None
+
+# 🔹 Function to Search FAISS
 def search_policy(query, top_k=3):
     if index is None:
         return "Error: FAISS index is not available."
@@ -61,16 +120,11 @@ def search_policy(query, top_k=3):
     return "\n\n".join([chunks[i] for i in indices[0] if i < len(chunks)])
 
 # 🔹 Hugging Face LLM Client
+from huggingface_hub import InferenceClient
 client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
 
-def respond(
-    message,
-    history: list[tuple[str, str]],
-    system_message,
-    max_tokens,
-    temperature,
-    top_p,
-):
+# 🔹 Function to Handle Chat Responses
+def respond(message, history, system_message, max_tokens, temperature, top_p):
     messages = [{"role": "system", "content": system_message}]
 
     for val in history:
@@ -113,11 +167,13 @@ def respond(
         yield response
 
 # 🔹 Gradio Chat Interface
+import gradio as gr
+
 demo = gr.ChatInterface(
     respond,
     additional_inputs=[
         gr.Textbox(
-            value="You are a knowledgeable and professional chatbot designed to assist Colorado case workers in determining eligibility for public assistance programs. Your primary role is to provide accurate, up-to-date, and policy-compliant information on Medicaid, SNAP, TANF, CHP+, and other state and federal assistance programs. Responses should be clear, concise, and structured based on eligibility criteria, income limits, deductions, federal poverty level guidelines, and program-specific requirements.",
+            value="You are a knowledgeable and professional chatbot designed to assist Colorado case workers in determining eligibility for public assistance programs.",
             label="System message"
         ),
         gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),