update model download sequence

.gitignore

@@ -0,0 +1,9 @@
+# Python
+__pycache__/
+*.py[cod]
+*$py.class
+*.so
+.Python
+env/
+venv/
+ENV/

app.py

@@ -3,9 +3,12 @@ CV Chatbot with RAG (Retrieval-Augmented Generation)
 FastAPI backend that uses semantic search to answer questions about your CV
 """
 
+import importlib
 import json
 import os
 import re
+import subprocess
+import sys
 import threading
 import time
 from typing import List, Dict, Optional, Tuple
@@ -94,6 +97,12 @@ from config import (
     SYSTEM_PROMPT
 )
 
+LLAMA_CPP_WHEEL_INDEX = os.getenv(
+    "LLAMA_CPP_WHEEL_INDEX",
+    "https://abetlen.github.io/llama-cpp-python/whl/cpu",
+)
+LLAMA_CPP_VERSION = os.getenv("LLAMA_CPP_VERSION", "0.3.16")
+
 # Initialize FastAPI
 app = FastAPI(title="CV Chatbot RAG API")
 
@@ -128,6 +137,39 @@ local_model_lock = threading.Lock()
 _session_serializer: Optional[URLSafeTimedSerializer] = None
 
 
+def ensure_llama_cpp_installed() -> None:
+    """Install the prebuilt CPU wheel for llama-cpp-python when needed."""
+    if LLM_PROVIDER != "local":
+        return
+
+    try:
+        importlib.import_module("llama_cpp")
+        return
+    except ImportError:
+        pass
+
+    package_spec = f"llama-cpp-python=={LLAMA_CPP_VERSION}"
+    print(f"Installing {package_spec} from llama.cpp CPU wheel index...")
+    cmd = [
+        sys.executable,
+        "-m",
+        "pip",
+        "install",
+        "--no-cache-dir",
+    ]
+    index_url = LLAMA_CPP_WHEEL_INDEX.strip()
+    if index_url:
+        cmd.extend(["--extra-index-url", index_url])
+    cmd.append(package_spec)
+
+    try:
+        subprocess.run(cmd, check=True)
+    except subprocess.CalledProcessError as install_err:
+        raise RuntimeError(
+            f"Failed to install {package_spec} from {index_url or 'PyPI'}"
+        ) from install_err
+
+
 def get_session_serializer() -> URLSafeTimedSerializer:
     """Lazily initialize the session token serializer."""
     global _session_serializer
@@ -356,11 +398,12 @@ def initialize_llm():
             raise ValueError("HUGGINGFACE_API_KEY not set in environment variables")
         print(f"Initialized HuggingFace Inference API with model: {HUGGINGFACE_MODEL}")
     elif LLM_PROVIDER == "local":
+        ensure_llama_cpp_installed()
         try:
             from llama_cpp import Llama  # type: ignore[import]
         except ImportError as import_err:
             raise ValueError(
-                "llama-cpp-python is not installed. Ensure requirements are up to date."
+                "llama-cpp-python could not be imported even after attempting installation."
             ) from import_err
 
         auth_token = LOCAL_MODEL_HF_TOKEN or None

requirements.txt

@@ -1,4 +1,3 @@
---extra-index-url https://download.pytorch.org/whl/cpu
 fastapi==0.104.1
 uvicorn[standard]==0.24.0
 sentence-transformers==2.2.2
@@ -7,7 +6,5 @@ faiss-cpu==1.8.0
 pydantic==2.5.0
 numpy>=1.26.4,<2
 python-multipart==0.0.6
-llama-cpp-python>=0.3.6
-torch==2.3.1
 itsdangerous==2.2.0
-requests==2.31.0
+requests==2.31.0