Spark2.5 inference support for MLX and MLX LM, enabling direct Hugging Face/MLX checkpoint loading, Apple Silicon GPU acceleration, and C++ extensions.
-
Updated
Oct 21, 2026 - Python
Spark2.5 inference support for MLX and MLX LM, enabling direct Hugging Face/MLX checkpoint loading, Apple Silicon GPU acceleration, and C++ extensions.
Apple Silicon performance monitor — asitop with M5 Max support (fixes M-cluster crash on M5/M5 Pro/M5 Max/M5 Ultra)
Measured microarchitecture tables for the Apple M5 CPU (P and E cores) and the rootless tool that produced them: ~940 instruction forms, ROB, rename, load/store queues, TLB
SuperMimic trained from scratch on Apple M5: MPS learning + CPU MuJoCo, all 40 LAFAN1 clips, reproducible setup, checkpoints, videos and measured results.
Native DisplayPort MST research for Apple Silicon M5: offline AUX/MST decoding, display-stack evidence, and receive-stage modeling. Research only; no working MST driver or qualified hardware.
Unlimited 16-byte data watchpoints and memory-access tracing for Apple Silicon Macs with MTE (M5+): find who writes a field across thousands of heap objects
To associate your repository with the apple-m5 topic, visit your repo's landing page and select "manage topics."