Skip to content

v0.10.0 - Updated Model Registry

Choose a tag to compare

@lucasventurasc lucasventurasc released this 11 Jan 19:25
· 84 commits to main since this release

What's New

Updated Model Downloads

All model download URLs now point to the official HuggingFace repository with Unsloth Dynamic (UD) quantizations:

Model Size VRAM Description
Loreguard Vanilla UD Q6_K 6.6 GB 8GB+ Recommended - best quality/size balance
Loreguard Vanilla UD Q5_K_M 5.7 GB 6-8GB Good quality, smaller size
Loreguard Vanilla UD Q4_K_M 4.9 GB 6GB Best for memory-constrained systems
Loreguard Vanilla UD Q8_0 8.5 GB 12GB+ Maximum quality

Changes

  • Q6_K is now the recommended default - Best balance of quality and VRAM usage
  • Added Q8_0 quantization - For users who want maximum quality
  • Fixed download URLs - All models now download from beyond-logic-labs/loreguard-vanilla-gguf
  • Removed MLX models - Will be added in a future release

Download

Models are automatically downloaded when you run the CLI:

loreguard --token YOUR_TOKEN

Or download manually:

huggingface-cli download beyond-logic-labs/loreguard-vanilla-gguf \
  loreguard-vanilla-UD-Q6_K.gguf --local-dir ./models

Full Changelog: v0.9.1...v0.10.0