The llama.cpp project has released version b12027, which enables tensor splitting for LFM3 models. The release also provides prebuilt downloads for platforms including macOS, iOS and several Ubuntu configurations.
The release notes do not explain which LFM3 models or hardware configurations are supported by the change.
Why it matters
Developers running LFM3 with llama.cpp can try tensor splitting, though the release notes do not specify the supported configurations or expected benefits.