A new nightly version has been released! ![]()
See the quickstart guide for installation instructions: Quickstart | Modular
MAX changelog updates:
-
The eager interpreter precompiles its matmul and unary-elementwise
graph-compiler models for the full(device, dtype)matrix at import by
default. SetMAX_EAGER_OP_PRECOMPILE=0to skip the import-time sweep and
compile each target lazily on first dispatch instead, bounding compile cost to
the targets a program actually uses. -
Added
max.graph.default_custom_extensions()and the
default_custom_extensions_scope()context manager. Paths registered as
defaults are merged into thecustom_extensionsof every newGraph, so a
backend can make its custom-op kernel library reachable from graphs built
without an explicitcustom_extensions=— including the eager-realization
graph that backsmax.experimentaltensors. Empty by default. -
Apple silicon GPU support for running MAX models has been extended to M1 and
M2 systems. Previously, the optimized matrix multiplication kernels for Apple
silicon GPUs only returned correct results on M3 and newer systems. That has
now been fixed for M1 and M2 systems, allowing many common MAX models to run
correctly on them.
Mojo changelog updates:
- [stdlib] Add
raise_python_exceptionerror helper - [stdlib] Make
DequeconditionallyImplicitlyDeletable - [stdlib] Rename
Reflected.field_typetoReflected.field - [Stdlib] Stabilize lifecycle traits
- [mojo-lang] Reject struct fields that expose
UnsafeAnyOrigin
Raw MAX diff: https://github.com/modular/modular/compare/12beae6f1434ac17dad9bd4b2947505c9f942296...4821f966f2c9beb67b7d29fe75ae808721687734)>
Current Mojo changelog: https://github.com/modular/modular/blob/main/mojo/docs/nightly-changelog.md