GGUF Editor D++

Version Windows Language License

🖥️ Desktop Application for high-performance GGUF model editing, tensor visualization, quantization, and multi-model merging. Built with Delphi for speed, reliability, and low memory footprint.

📖 Overview

GGUF Editor D++ is a feature-rich, native Windows application designed for AI researchers, quantization engineers, and model developers. It provides a seamless workflow for inspecting, comparing, converting, and editing GGUF and Safetensors models. Leveraging memory-mapped I/O, parallel processing, and optimized C/DLL delegates, it handles large-scale models (70B+) with minimal RAM usage and maximum responsiveness.

✨ Key Features

Category Capabilities
🔍 Tensor Visualization Real-time charts, histograms, and difference statistics (RMS, MAE, Mean). Compare Model A vs B with outlier/block highlighting.
🔄 Quantization Engine Full support for GGML quantization types. Custom presets, RMSE optimization, and parallel chunk-based conversion.
📊 Multi-Model Comparison Load GGUF (Model A/B) + Safetensors side-by-side. Visualize shape/dtype mismatches and absent tensors.
🏷️ Smart Mapping & Filtering Apply regex-style tensor naming patterns, ignore prefixes, and filter by layer/type dynamically.
✏️ Metadata & KV Editor Full control over GGUF key-value pairs. Edit strings, arrays, floats, and convert types on-the-fly.
📦 Split & Merge Shard large models or merge multiple sources into a single aligned GGUF file. Supports Safetensors sharding.
⚡ Performance Optimized MMAP file reading, System.Threading parallel quantization, DLL delegation for heavy math, and zero-copy buffer handling.
🌍 Multi-Language UI 10+ languages supported with dynamic runtime translation.
💾 Persistent Configuration Auto-saves window positions, quant presets, filters, and UI preferences.
Main App

Screenshot Main App.

🛠️ Installation & Requirements

System Requirements

Dependencies

  1. GGUF_Editor.exe (Main executable)
  2. ggml-base.dll (Required for optimized quantization/dequantization backends)
    • Place it in the same directory as the executable.
    • Provides C-level delegates for dequantize_row_* and quantize_* routines.
  3. Standard VCL/TeeChart runtime (bundled with release)

⚠️ Note: This is a desktop application, not a machine learning model. It runs locally on your machine and does not upload data.

📊 Supported Formats & Quantizations

Input Formats

Quantization Support

Standard K-Quant IQ Series Ternary FP4 / MoE
Q4_0, Q4_1, Q5_0, Q5_1, Q8_0, Q8_1 Q2_K to Q8_K IQ2_XXS, IQ2_XS, IQ2_S, IQ3_XXS, IQ3_S, IQ1_S, IQ1_M, IQ4_NL, IQ4_XS TQ1_0, TQ2_0 MXFP4, NVFP4
F16, BF16, F32, F64, I8/I16/I32/I64

🚀 Quick Start

  1. Download the latest release from the repository.
  2. Extract the folder and ensure ggml-base.dll is present.
  3. Launch GGUF_Editor.exe.
  4. Load Models:
    • Drag & drop GGUF/Safetensors files into the Model A, Model B, or Safetensors tabs.
    • Or use File > Open menus.
  5. Compare & Edit:
    • Select tensors in the list view.
    • Apply quantization via the DType dropdown.
    • Visualize differences in the View Tensors tab.
    • Edit metadata in Tools > Edit KV Pairs.
  6. Save/Export: Configure split size, output path, and click Save.
workflow_diagram

Workflow diagram.

🤝 Credits & Support

📜 License

This project is released under the MIT License. You are free to use, modify, and distribute it for personal or commercial purposes. See LICENSE for full terms.

🐛 Known Issues & Notes

⭐ If you find this tool useful, please star this repository and consider supporting development on Ko-fi!