Optimize MLX for Apple Silicon M5 Max#3356

Closed

ambermontlabs wants to merge 24 commits intoml-explore:mainfrom

ambermontlabs:feature/m5-max-optimizations

Commits on Apr 3, 2026

Optimize MLX for Apple Silicon M5 Max
ambermontlabs
committed

Commits on Apr 4, 2026

M5 Max Performance Optimizations
ambermontlabs
committed
docs: Add benchmarking documentation and helper script
ambermontlabs
committed
docs: Add Metal build prerequisites and troubleshooting
ambermontlabs
committed
docs: Clarify Xcode requirement for Metal compiler
ambermontlabs
committed
docs: Add Metal Toolchain download step
ambermontlabs
committed
docs: Add Metal Toolchain troubleshooting and verification steps
ambermontlabs
committed
bench: Add helpful error messages for MLX import failures
ambermontlabs
committed
fix: Use relative import for time_utils
ambermontlabs
committed
fix: Support both direct script execution and module import for time_utils
ambermontlabs
committed
Create build.log
ambermontlabs
committed
Merge branch 'feature/m5-max-optimizations' of https://github.com/ambermontlabs/mlx into feature/m5-max-optimizations
ambermontlabs
committed
fix: Use mx.default_device() instead of mx.device()
ambermontlabs
committed
fix: Use CPU stream for unsupported GPU linalg operations
ambermontlabs
committed
fix: Remove unsupported matrix_power linalg operation
ambermontlabs
committed
feat: Add M5 Max performance optimizations
ambermontlabs
committed
fix: Add missing math import
ambermontlabs
committed
fixes
ambermontlabs
committed
fix: Replace mx.nn with correct API functions
ambermontlabs
committed
fix: Remove duplicate softmax function and fix class calls
ambermontlabs
committed
fix: Rename benchmark_softmax_reduce to benchmark_softmax
ambermontlabs
committed
fix: Change rms_norm to mx.fast.rms_norm
ambermontlabs
committed
benchmark M5 Max
ambermontlabs
committed
docs: performance benchmarks on M5 Max
ambermontlabs
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Optimize MLX for Apple Silicon M5 Max#3356

Optimize MLX for Apple Silicon M5 Max#3356
ambermontlabs wants to merge 24 commits intoml-explore:mainfrom
ambermontlabs:feature/m5-max-optimizations

Commits on Apr 3, 2026

Optimize MLX for Apple Silicon M5 Max

Commits on Apr 4, 2026

M5 Max Performance Optimizations

docs: Add benchmarking documentation and helper script

docs: Add Metal build prerequisites and troubleshooting

docs: Clarify Xcode requirement for Metal compiler

docs: Add Metal Toolchain download step

docs: Add Metal Toolchain troubleshooting and verification steps

bench: Add helpful error messages for MLX import failures

fix: Use relative import for time_utils

fix: Support both direct script execution and module import for time_utils

Create build.log

Merge branch 'feature/m5-max-optimizations' of https://github.com/ambermontlabs/mlx into feature/m5-max-optimizations

fix: Use mx.default_device() instead of mx.device()

fix: Use CPU stream for unsupported GPU linalg operations

fix: Remove unsupported matrix_power linalg operation

feat: Add M5 Max performance optimizations

fix: Add missing math import

fixes

fix: Replace mx.nn with correct API functions

fix: Remove duplicate softmax function and fix class calls

fix: Rename benchmark_softmax_reduce to benchmark_softmax

fix: Change rms_norm to mx.fast.rms_norm

benchmark M5 Max

docs: performance benchmarks on M5 Max