Unsloth
Named by 18 kept items in this workspace.
15 items name it under frameworks and tools and 3 under classification.technologies.
The items that name it 18
Releases dynamic GGUF quantisations, here DeepSeek V4.1-Flash-1017 Dynamic v3.1, and benchmarked them against GSQ-RCO.
Provides a 1-bit GGUF of Mistral Large 4 at 223GB versus 1.1TB in FP8, retaining 91% top-1 accuracy and running on a Mac Studio.
Provides quantised model builds such as a 2-bit MiniMax M3.1 Flash.
Provides dynamic GGUF quantizations and Unsloth Desktop for local running of models.
Shipped dynamic GGUFs of DeepSeek V4.1 Flash 1017 for local deployment.
Published quants of Muse Spark Open, UD-IQ1_S at 176GB
Fine-tuning Qwen3.5-0.8B for the decision model
Free notebook to turn Qwen3.5-4B into a decision model on 8GB VRAM
Added OS-level sandboxing with under 100ms overhead per tool call
Fine-tuning/quantisation tool providing day-0 support and GGUF conversions for EmbeddingGemma 2
LoRA training framework built on transformers, could train LoRA directly over GGUF models
LoRA training framework that could use native GGUF support in transformers for lower-memory training
Quantization/finetuning toolkit used to produce the UD-Q4_K_XL quant of Qwen3.8-27B
Optional backend (backend: unsloth) providing 2-5x faster training, installed via pip install soup-cli[fast]
Supported inference deployment framework
In every workspace you can read 19
AI
All technologies in this workspace.