install.json

{
  "active_models": ["Qwen3.5-4B"],
  "model_dir": "./model",
  "models": {
    "Qwen3-1.7B": {
      "repo": "unsloth/Qwen3-1.7B-GGUF",
      "file": "Qwen3-1.7B-Q4_K_M.gguf",
      "note": "~1.0GB — fastest, default"
    },
    "Qwen3.5-4B": {
      "repo": "unsloth/Qwen3.5-4B-GGUF",
      "file": "Qwen3.5-4B-IQ4_NL.gguf",
      "note": "~2.6GB — good balance"
    },
    "DeepSeek-R1-8B": {
      "repo": "unsloth/DeepSeek-R1-Distill-Llama-8B-GGUF",
      "file": "DeepSeek-R1-Distill-Llama-8B-Q4_K_M.gguf",
      "note": "~4.6GB — deep reasoning / R1 distill"
    },
    "Llama-3.1-8B": {
      "repo": "unsloth/Llama-3.1-8B-Instruct-GGUF",
      "file": "Llama-3.1-8B-Instruct-Q4_K_M.gguf",
      "note": "~4.6GB — general purpose"
    },
    "Phi-4-mini": {
      "repo": "unsloth/Phi-4-mini-reasoning-GGUF",
      "file": "Phi-4-mini-reasoning-Q4_K_M.gguf",
      "note": "~2.3GB — math / step-by-step reasoning"
    }
  },
  "chunks": {
    "chunk_mb": 100
  },
  "server": {
    "port": 5000,
    "max_new_tokens": 1500,
    "temperature": 0.6,
    "top_p": 0.9,
    "max_seq_len": 4096,
    "n_batch": 1024,
    "use_mlock": true
  },
  "needed": [
    "flask",
    "huggingface-hub",
    "numpy",
    "chromadb",
    "sentence-transformers",
    "llama-cpp-python"
  ]
}