Launches
BoxyFit says its new Groot pack runs on a 12GB GPU — with a runtime catch
Community reports 4 sources
G
Groot Gyro-M
Model Nov 3, 2026
BoxyFit's community TQ2_T quantization of Groot-65B-A6B, using the Strata runtime and an expert cache to target consumer GPUs.
Visit primary link
2 sources attached
What happened lately
Groot Gyro-M timeline
Groot Gyro-M is a community quantization of the existing Groot-65B-A6B-261023 model. BoxyFit dates this combination to November 3, 2026. It uses TQ2_T expert weights and requires the Strata runtime; upstream llama.cpp cannot load the file despite its GGUF extension and Q2_0 filename label.
The publisher’s smaller-memory fit table was measured on an RTX 3090 with capped expert caches to emulate smaller cards. Its speed and quality figures are author-reported. The pack is listed under Qwen Community 1.0 terms, separately from Strata’s MIT runtime license. The documented recipe requires CUDA architecture 86 or newer, Strata 0.2.8 or later and engine 0.1.54, with fast NVMe storage for the large host-side lookup table.