v1.4.0 (prerel)

AutoGGUF v1.4.0 prerelease

Changelog:

• LoRA Conversion Section: Added a new section for converting HuggingFace PEFT LoRA adapters to GGML or GGUF formats.
• Output Type Selection: Added a dropdown menu to choose between GGML and GGUF output formats for LoRA conversion.
• Base Model Selection: Added a folder selection dialog to choose the base model (in safetensors format) when GGUF output is selected.
• LoRA Adapter List: Added a list widget to add multiple LoRA adapters with individual scaling factors.
• Export LoRA Section: Added a section to merge LoRA adapters into a base model, supporting multiple adapters and scaling.
• Task Names: Updated the task names in the task list to be more informative, including the input and output file names.
• IMatrix Generation Check: Added a check to prevent IMatrix generation from starting if the model path is empty.

Currently, this build contains an src folder with the conversion tools. In the future I might simply add a feature to download the conversion scripts from the llama.cpp GitHub, but they are currently hardcoded here.

Planned features in next release:

• exe favicon

v1.3.1

AutoGGUF v1.3.1

Changelog:

• Added AUTOGGUF_CHECK_BACKEND environment variable to disable backend check on start (enabled/disabled)
• --onefile build with PyInstaller, _internal directory is no longer required

Full Changelog: v1.3.0...v1.3.1

v1.3.0

AutoGGUF v1.3.0

Changelog:

• Added support for new llama-imatrix parameters:
  • Implemented context size (--ctx-size) input
  • Added threads (--threads) control
• Added new parameters to IMatrix section layout for better usability:
  • Converted context size input to a QSpinBox for better control
  • Implemented a slider-spinbox combination for thread count selection
  • Changed output frequency input to a QSpinBox with a range of 1-100 and percentage suffix
• Remove duplicated functions
• Fixed error when loading presets containing KV overrides
• Updated generate_imatrix() method to use new UI element values
• Improved error handling in preset loading
• Enhanced localization support for new UI elements (fallback to English)
• Added new localization keys for:
  • Context size
  • Thread count
  • Updated IMatrix generation related strings
• Adjusted default values and ranges for new numeric inputs, input types
• Improved code readability and maintainability

Full Changelog: v1.2.1...v1.3.0

v1.2.1

AutoGGUF v1.2.1

Changelog:

• Added Refresh Models button
• Fixed iostream llama.cpp issue, quantized_models directory created on launch
• Added Linux build (built on Ubuntu 24.04 LTS)

Full Changelog: v1.2.0...v1.2.1

v1.2.0

AutoGGUF v1.2.0

Changelog:

• Added more robust logging, find them at latest_<timestamp>.log in your logs folder
• Added localizations with support for 28 languages (machine translated using Gemini Experimental 0801)

Full Changelog: v1.1.0...v1.2.0

v1.1.0

AutoGGUF v1.1.0

Changelog:

New Features:

• Added support for multiple KV overrides for more flexible model metadata customization
• Improved CUDA checking ability and extraction to the backend folder

UI Improvements:

• Implemented a scrollable area for KV overrides with add/delete capabilities
• Enhanced the visibility and usability of the Output Tensor Type and Token Embedding Type options

Bug Fixes:

• Corrected behavior of Output Tensor Type and Token Embedding Type dropdown menus
• Resolved various minor UI inconsistencies and improved error handling

Other Changes:

• Refactored code for better modularity and reduced circular dependencies
• Improved error messages and user feedback

Full Changelog: v1.0.1...v1.1.0

v1.0.1

Hotfix and binaries

Changelog:

• Added Windows binary (created using PyInstaller)
• Fixed issue where quantization errored with "AutoGGUF does not have x attribute"

Full Changelog: v1.0.0...v1.0.1

v1.0.0

AutoGGUF initial release

Changelog

Added:

• GUI interface for automated GGUF model quantization
• System resource monitoring (RAM and CPU usage)
• Llama.cpp backend selection and management
• Automatic download of llama.cpp releases from GitHub
• Model selection from local directory
• Comprehensive quantization options:
  • Quantization type selection
  • Allow requantize option
  • Leave output tensor option
  • Pure quantization option
  • IMatrix support
  • Include/exclude weights options
  • Output tensor type selection
  • Token embedding type selection
  • Keep split option
  • Override KV option
• Task list for managing multiple quantization jobs
• Real-time log viewing for quantization tasks
• IMatrix generation feature with customizable settings
• GPU offload settings for IMatrix generation
• Context menu for task management (Properties, Cancel, Retry, Delete)
• Detailed model information dialog
• Error handling and user notifications
• Confirmation dialogs for task deletion and application exit

Full Changelog: https://github.com/leafspark/AutoGGUF/commits/v1.0.0