gguf-merge-tool 9.2.0

Professional tool for downloading, verifying and merging sharded GGUF models

GGUF Merge Tool v9.2

Professional GUI + CLI tool for downloading, verifying and merging sharded GGUF models from Hugging Face. Table of Contents

Features

Quick Start

Installation

CLI Usage

GUI Usage

Configuration

Requirements

Troubleshooting

Performance Tips

Technical Details

License

Acknowledgments

Support

Features

Modern GUI - Tkinter-based interface with scrollable file list and mouse wheel support

Smart Grouping - Files automatically grouped by quantization type (Q4_K_M, Q5_K_S, etc.)

Resume Downloads - HTTP Range requests with smart fallback protection (200 vs 206 handling)

Parallel Downloading - Configurable threads (auto-detects CPU cores)

Hash Verification - SHA256 verification using HF LFS metadata

Smart Merge - Uses llama-gguf-split from llama.cpp

CLI Support - download and merge subcommands for automation

Smart Detection - Automatically finds llama-gguf-split in common locations

Path Persistence - Saves llama.cpp path between sessions

Retry Logic - Automatic retry with exponential backoff (3 attempts)

Real-time ETA - Shows speed (MB/s) and estimated time remaining

Multi-language - English, Russian, Chinese, Spanish, German, Japanese documentation Quick Start Windows

Install dependencies: pip install huggingface_hub tqdm requests

Download llama-gguf-split from: https://github.com/ggerganov/llama.cpp/releases

Run the tool: python gguf_merge_tool.py Linux / macOS

Install dependencies: pip install huggingface_hub tqdm requests

Run the tool: python3 gguf_merge_tool.py Installation From Source (Recommended)

Clone the repository: git clone https://github.com/yourusername/gguf-merge-tool.git cd gguf-merge-tool

Install dependencies: pip install -r requirements.txt

Run the tool: python src/gguf_merge_tool.py From PyPI (Coming Soon)

pip install gguf-merge-tool gguf-merge-gui Windows Users: Required Binary

Download llama-gguf-split.exe from llama.cpp releases Place it in one of these locations:

tools/llama-gguf-split.exe

bin/llama-gguf-split.exe

Or specify the path in GUI settings

CLI Usage Download Specific Files

python src/gguf_merge_tool.py download -r unsloth/Qwen3.5-122B-A10B-GGUF -d ./models -f Qwen3.5-122B-A10B-Q4_K_M-00001-of-00008.gguf Qwen3.5-122B-A10B-Q4_K_M-00002-of-00008.gguf Download Entire Repository

python src/gguf_merge_tool.py download -r unsloth/Qwen3.5-122B-A10B-GGUF -d ./models Download with Authentication (Private Repo)

python src/gguf_merge_tool.py download -r your-username/your-private-model -d ./models -t hf_xxxxxxxxxxxxxxxxxx Merge Parts

python src/gguf_merge_tool.py merge -d ./models -o ./models/merged_model.gguf -f CLI Arguments Reference

Command: download -d, --dir - Directory to save files - Required: Yes -r, --repo - Hugging Face repository ID - Required: Yes -b, --branch - Repository branch (default: main) - Required: No -t, --token - HF access token - Required: No -f, --files - List of files to download - Required: No

Command: merge -d, --dir - Directory with parts - Required: Yes -o, --output - Output file path - Required: No -f, --force - Overwrite existing file - Required: No GUI Usage Step 1: Load Repository

Enter Hugging Face repo ID (e.g., unsloth/Qwen3.5-122B-A10B-GGUF) Enter branch (default: main) Enter your HF token (required for private repos) Click "Load" button Step 2: Select Files

Files are automatically grouped by quantization type Click "Select All" to select all files Or select individual files with checkboxes Step 3: Download

Click "Download" or "Download All" Monitor progress in real-time:

Progress bar shows percentage

Status bar shows: [1/4] filename.gguf: 45% | 4.5 MB/s | ETA: 2h 15m

Step 4: Verify (Optional but Recommended)

Click "Verify" button Tool validates SHA256 hashes against HF metadata Shows success message if all hashes match Step 5: Merge (Optional)

Select at least 2 parts (checkboxes) Click "Merge" button Tool calls llama-gguf-split --merge Output file is saved with correct base name Configuration llama.cpp Path

In GUI, locate "Path to llama.cpp": Click "Find" for auto-detection Or click "Browse" to select manually Path is saved to llama_path.txt for persistence Thread Settings

Adjust JOBS constant in code (default: min(4, os.cpu_count() or 2)): In gguf_merge_tool.py JOBS = min(4, os.cpu_count() or 2) - Change this value Disk Space Margin

Change DISK_SPACE_MARGIN (default: 1.1 = 10% extra): DISK_SPACE_MARGIN = 1.1 Download Chunk Size

Change CHUNK_SIZE (default: 1 MB): CHUNK_SIZE = 1024 * 1024 Requirements

Python >= 3.8 huggingface_hub >= 0.20.0 tqdm >= 4.60.0 requests >= 2.28.0 gguf >= 0.1.0 (optional, for metadata reading) llama-gguf-split (from llama.cpp) Verify Installation

python -c "import huggingface_hub, tqdm, requests; print('All dependencies installed')" Troubleshooting llama-gguf-split not found

Download from llama.cpp releases Place in tools/ folder or specify path in GUI HTTP 401 Unauthorized or HTTP 403 Forbidden

Enter your Hugging Face token Ensure token has read access to the repository Token format: hf_xxxxxxxxxxxxxxxxxx Insufficient disk space

Free up disk space (need 10% extra for merge) Change download directory to another drive Use DISK_SPACE_MARGIN to adjust buffer Download stuck at 0%

Check internet connection Verify token is correct Check if repository is private Check if repository exists File corrupted after resume

Tool has built-in Range fallback protection If server doesn't support Range (returns 200 instead of 206), it starts download from scratch This is intentional to prevent file corruption GUI doesn't start

Check Tkinter is installed (usually included with Python) On Linux: sudo apt-get install python3-tk On macOS: brew install python-tk Performance Tips

Many small files - Keep default threads (4) Few large files - Increase threads to 8 Slow network - Reduce threads to 2 Fast NVMe SSD - Keep default or increase Network interruptions - Tool handles resume automatically Technical Details Download Flow

Parse HF repo to get file list with sizes and hashes User selects files to start download Check disk space - abort if insufficient Parallel download with Resume (Range requests) Smart Range fallback (200 vs 206 handling) Real-time progress with speed calculation SHA256 verification after download Merge via llama-gguf-split ETA Calculation

ETA = (Total Bytes - Downloaded Bytes) / Current Speed Supported Quantizations

F16, F32, Q8_0, Q6_K, Q5_K_M, Q5_K_S, Q4_K_M, Q4_K_S, Q3_K_M, Q3_K_S, Q2_K, IQ4_XS, IQ3_XS, IQ2_XS, IQ1_S License

MIT License - see LICENSE file for details. Acknowledgments

llama.cpp for the merge utility Hugging Face for the model hosting Python Tkinter for GUI Support

Issues: GitHub Issues Discussions: GitHub Discussions Star This Project

If you find this tool useful, please give it a star on GitHub!

Made with ❤️ for the AI community