MAX Nightly 25.4.0.dev2025050705 Released

Nightly
1

:astronaut: A new nightly version has been released! :astronaut:

Update to the latest nightly using Magic via
magic init --format mojoproject -c conda-forge -c https://conda.modular.com/max-nightly/
or add https://conda.modular.com/max-nightly/ as a channel to mojoproject.toml

MAX changelog updates:

  • GGUF quantized Llamas (q4_0, q4_k, and q6_k) are now supported with paged
    KVCache strategy.

  • The naive KVCache has been deleted.

  • Added fold op for combining sliding blocks into a larger tensor.

Mojo changelog updates:

  • [tooling] Include associated aliases in trait docs

Raw MAX diff: https://github.com/modular/modular/compare/ea12bdd7f384b8726844b2da2575058f189c09f5...a50af073ccdb55579b495e7230b1554595916b77
Current Mojo changelog: https://github.com/modular/modular/blob/main/mojo/docs/changelog.md