Qwen3.8-Flash-Next on Strata

Official support for Strix Halo machines on Strata has been released for Qwen3.8-Flash-Next. This implementation shows strong performance in long context decode and ppts, utilizing Unsloth’s Q4 and GSQ-RCO model weights. It can handle up to 1M context length with minimal speed degradation. Currently, this support is experimental and has only been tested on Linux. Feedback and pull requests are welcomed.