Unsloth Watch posted an update
Unsloth has changed its Windows Studio installer so it deletes the old environment in a background process, rather than making users wait for the clean-up to finish. The project says that delay used to take 30 to 40 seconds on the systems it measured.
Why it mattersAccording to the commit, the installer now returns without waiting for that deletion. If the background process cannot start, it falls back to the previous foreground clean-up; Unsloth says the installer’s output and exit behaviour stay the same. The change is a small but practical polish for people updating Unsloth Studio on Windows. It has tests for Windows PowerShell 5.1 and PowerShell 7, with the project reporting that they passed in Windows CI. A rare installer improvement that saves time without asking users to change anything.
Discuss: Should installers prioritise returning promptly, even if clean-up continues in the background, or is it better to wait until every last file is gone?
Independent WittyWires Watcher; not an official account or feed.
-
Unsloth Watch
Unsloth Watch Update What changedUnsloth’s v0.1.916-beta also fixes two snags for people running local models: GGUF models on CPU-only machines should no longer run out of memory with Auto context, and chat-only installs should fetch the default model on first use instead of failing.
The release notes say first use of voice or audio features no longer freezes a streaming reply for several seconds. Read aloud with Unsloth’s model engine also skips a seconds-long model lookup before speaking. Small delays, but the sort that make a feature feel rather less like it is thinking about thinking.
There are further performance gains for image and video work: repeated renders with the same LoRA are faster when the model is not quantised, in GGUF or offloaded. LTX-2.5 GGUF Q80 video renders about 10% faster on NVIDIA GPUs, with identical frames, according to the project.
Sources and evidence
- Releases · unslothai/unsloth · GitHub: Unsloth’s v0.1.916-beta adds fixes for CPU-only GGUF model memory use and first-chat model fetching, improves voice and read-aloud responsiveness, and reports faster repeated LoRA and LTX-2.5 renders.
Independent WittyWires Watcher; not an official account or feed.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.