Phi-4-mini GGUF Outfit Recommendation

Outfit recommendation inference using unsloth/Phi-4-mini-instruct-GGUF model with llama-cpp-python.

Setup

# Install uv if not already
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install dependencies
uv sync

Download Model

uv run python download_model.py --filename Phi-4-mini-instruct-Q4_K_M.gguf

Available quantizations:

• Q2_K.gguf (1.68 GB)
• Q3_K_M.gguf (2.12 GB)
• Q4_K_M.gguf (2.49 GB) - default
• Q5_K_M.gguf (2.85 GB)
• Q6_K.gguf (3.16 GB)
• Q8_0.gguf (4.08 GB)

Usage

Run with fixed input values

uv run python run_inference.py

Run with custom parameters

uv run python inference.py \
    --country USA \
    --height 170 \
    --weight 65 \
    --occasion "Business Casual" \
    --time "Afternoon"

GPU acceleration (Linux)

CMAKE_ARGS="-DGGML_CUDA=on" uv sync
uv run python inference.py --n_gpu_layers 32

Input Parameters

Parameter	Description	Example
STYLE_COUNTRY	Country for style preferences	USA
STYLE_HEIGHT	Height in cm	170
STYLE_WEIGHT	Weight in kg	65
STYLE_OCCASION	Occasion type	Business Casual
STYLE_TIME	Time of day	Afternoon

Output Schema

{
  "complete_outfits": [
    {
      "name": "string",
      "description": "string",
      "occasion_type": "string",
      "season": ["Spring", "Summer", "Autumn", "Winter"],
      "body_shape": ["Hourglass", "Apple", "Pear", "Rectangle", "Inverted Triangle"],
      "top": {
        "category": "tshirt | shirt | blouse | ...",
        "name": "string",
        "description": "string",
        "color": "string",
        "hex_code": "#RRGGBB",
        "fabric": "string",
        "fit": "slim | regular | relaxed | oversized | tailored",
        "neckline": "string",
        "sleeve_length": "short | 3/4 | long | extra_long",
        "pattern": "string or null",
        "occasion_suitability": ["..."],
        "season_suitability": ["..."],
        "body_shape_suitability": ["..."],
        "styling_tips": ["..."]
      },
      "bottom": { ... } or null,
      "footwear": { ... },
      "outerwear": { ... } or null,
      "accessories": [{ ... }] or null,
      "overall_color_palette": [{ "color_name": "...", "hex_code": "...", "reason": "..." }],
      "styling_tips": ["..."],
      "why_this_outfit": "string"
    }
  ]
}

Files

• inference.py - Full-featured CLI inference script
• run_inference.py - Simple script with fixed values
• download_model.py - Model downloader from HuggingFace
• config/config.py - Configuration settings
• prompts/input.txt - Input parameters and output schema
• pyproject.toml - Project dependencies (uv)