Guides
Pages for the jobs this engine does: which name to search, running Qwen on your PC, and which machines fit.
Strata here is the local inference engine at Niko1221/Strata. It runs Qwen3.8-Flash-Next on a Windows or Linux PC with an NVIDIA or AMD GPU. The same word is also used for rock layers and for some legal writing. If you want this project, search Niko Strata or Strata local.
Qwen3.8-Flash-Next is the model these pages cover. The weights stay on your machine. Pick a quant on the models page, then follow install. The smallest quants ask for about 12 GB of VRAM and 48 GB of RAM.
These benchmark rows have 16 GB of memory or less. A tight fit may need a smaller quant. Open a row for the measured read and write speeds. The list comes from the benchmark file. It is not a promise that every 16 GB card runs every quant.
AMD rows from the same benchmark file: discrete Radeon cards and the Ryzen AI Max machine. Install covers the HIP build. Speeds are on each machine's page.