to run the project backend: set PYTHONPATH=%CD%\backend

frontend: cd frontend npm run dev

---

🚀 Setup & Installation

Prerequisites

• Python 3.11
• Git
• (Optional) OpenAI API key or local LLM (Ollama)

Installation Steps

1. Clone the repository

   git clone <repository-url>
   cd Code_IQ

2.create virtual environment conda create -n code_iq conda activate code_iq

3.Install dependencies pip install -r requirements.txt

4.Setup Local LLM (Option A: Direct llama-cpp-python - Recommended for GPU)

For direct local LLM inference with GPU acceleration:

Step 1: Install llama-cpp-python with GPU support

# For NVIDIA CUDA GPU (RTX, GTX, etc.)
pip install llama-cpp-python --upgrade

# If wheel build fails (missing C++ compiler), use conda instead:
conda install -c conda-forge llama-cpp-python -y

Step 2: Download GGUF model Download from HuggingFace and save to models/ folder: https://huggingface.co/unsloth/DeepSeek-R1-Distill-Qwen-1.5B-GGUF

Or download via Python:

from huggingface_hub import hf_hub_download
hf_hub_download(
    repo_id="unsloth/DeepSeek-R1-Distill-Qwen-1.5B-GGUF",
    filename="DeepSeek-R1-Distill-Qwen-1.5B-Q4_K_M.gguf",
    local_dir="models"
)

Step 3: Verify CUDA Setup

# Check NVIDIA GPU availability
nvidia-smi

# Verify in Python
python -c "from llama_cpp import Llama; print('llama-cpp-python installed!')"

Step 4: Test local inference

python test/local_llm.py

Configuration in config/llm.yaml:

local:
  enabled: true
  mode: "llama_cpp"
  model_path: "models/DeepSeek-R1-Distill-Qwen-1.5B-Q4_K_M.gguf"
  n_ctx: 8192
  n_gpu_layers: -1 # All layers on GPU
  n_threads: null # Auto-detect

Troubleshooting wheel build errors: If you get "No CMAKE_C_COMPILER could be found" error:

1. Option 1 (Recommended): Use conda pre-built wheels

   conda install -c conda-forge llama-cpp-python

2. Option 2: Install Visual Studio Build Tools with C++ workload
   • Download from: https://visualstudio.microsoft.com/downloads/
   • Select "Desktop development with C++"
   • Restart terminal after install

---

5.Setup Ollama (Option B: Alternative HTTP-based) Minimal setup (10 minutes) 1️⃣ Install Ollama

https://ollama.com/download/windows

2️⃣ Pull a model (do this once) ollama pull qwen2.5:7b

5.to run the project backend: set PYTHONPATH=%CD%\backend python -m uvicorn backend.app:app --reload --reload-dir backend

frontend: cd frontend npm run dev

Data folder location:

• Runtime data (cloned repos + intermediate outputs) is stored outside this app folder at: ../data
• Effective paths: ../data/input/repositories ../data/intermediate/agent_output ../data/intermediate/navigator_output

conda install -c conda-forge llama-cpp-python -y

IMP COMMANDS:

python -m uvicorn backend.app:app --reload

taskkill /F /IM python.exe

cd .. rmdir /s /q .cache