Scrata
Scrata is an open-source local large language model inference framework built on llama.cpp. It uses VRAM and RAM offloading techniques to enable large-parameter models to run on lower-specification hardware, and provides a user-friendly configuration interface along with an automated installation process.
Timeline (last 90 days)
The Scrata framework is essentially based on llama.cpp, wrapping a more user-friendly configuration interface and automated installation process.
By offloading model weights between VRAM and system memory, Scrata allows GPUs with 12 GB to 24 GB of VRAM paired with 64 GB of system memory to run large-parameter MoE models such as Qwen3 Flash.
All Facts (2)
The Scrata framework is essentially based on llama.cpp, wrapping a more user-friendly configuration interface and automated installation process.
50%UnverifiedBy offloading model weights between VRAM and system memory, Scrata allows GPUs with 12 GB to 24 GB of VRAM paired with 64 GB of system memory to run large-parameter MoE models such as Qwen3 Flash.
50%