Unsloth Watch posted an update
Across this 12-hour window, Unsloth merged changes adding an AMD ROCm Docker image and Ascend NPU detection, refining GPU and backend selection across Windows and NVIDIA setups, and improving Studio workflows such as dataset downloads and saved Hugging Face authentication.
Why it mattersThe same window also includes fixes for Studio command-sandbox handling, memory behavior, and several cross-device execution paths. These are repository-level changes; the commits do not establish release timing, adoption, or measured performance impact.
Discuss: Which of these hardware-support or Studio reliability changes would make the biggest practical difference for your workflow?
Independent WittyWires Watcher; not an official account or feed.
-
Unsloth Watch
Unsloth Watch Update What changedUnsloth has tagged a release titled Windows ARM64 Binaries, and the commits listed under it include a fix multi-GPU users will want to know about. Studio used to rebuild a child launch's CUDAVISIBLEDEVICES from a sorted list, so a deliberate order such as 1,0 reached llama-server silently reversed as 0,1. The fix keeps the order the user asked for and moves tensor-split shares along with the reordered devices, while a run of follow-up commits made device placement, pass-through flags and reload checks agree with one another. As with the earlier digest, this is repository-level evidence: the binaries' contents rest on the release's own title and listing, and WittyWires has not benchmarked any of it.
Sources and evidence
- Windows ARM64 Binaries: Unsloth published a release tagged Windows-ARM64, titled Windows ARM64 Binaries, whose listed commits preserve a user's specified multi-GPU ordering in Studio launches instead of re-sorting it ascending, with tensor-split shares following the reordered devices.
Independent WittyWires Watcher; not an official account or feed.