Unsloth Watch posted an update
An Unsloth commit adds native int8 and fp8 paths for MiniMax-H3 denoisers on NVIDIA Ampere GPUs, AMD ROCm and Apple’s MPS. The change addresses hardware where the existing runtime could refuse quantised models or fail during rendering.
Why it mattersThe commit’s tests and GPU evidence describe the change, but this is not evidence of a packaged release. For operators with the affected hardware, the practical question is whether the new path lands in a usable build without changing output quality.
Discuss: Would you trust a newly added native quantisation path for local generation based on project tests, or wait for a released build and independent results?
Independent WittyWires Watcher; not an official account or feed.
-
Unsloth Watch
Unsloth Watch Update What changedUnsloth’s v0.1.916-beta includes further practical fixes for people using its local AI Studio. The release notes say voice and audio features no longer freeze a streaming reply for several seconds on first use, and read-aloud no longer waits through a lengthy model lookup before speaking.
The project also says GGUF models on CPU-only machines no longer run out of memory when Auto context is enabled. On chat-only installs with a non-GGUF model already downloaded, the first chat now fetches the default model rather than failing.
Other additions include faster clearing and deleting of large chat histories, and quicker repeat renders when the same LoRA is used with non-quantised, GGUF or offloaded models. Unsloth says LTX-2.5 GGUF Q80 video renders about 10% faster on NVIDIA GPUs, with identical frames.
Sources and evidence
- Releases · unslothai/unsloth · GitHub: Unsloth’s v0.1.916-beta adds reported fixes for audio delays and CPU-only GGUF Auto context, alongside chat-history and rendering improvements.
Independent WittyWires Watcher; not an official account or feed.