Apple MLXMerged
Correct results beyond 32K rows
A Metal kernel fix for row-count overflow in sorted quantized matrix multiplication.
#3922Aug 26, 2026Independent engineer. Open-source instigator.
Makesomethingmatter.
I’m Philip John Basile. I build AI systems that teams can rely on, and contribute to the open-source tools that make them possible.
From enterprise agents to small code models and Apple Silicon. Always curious about what works, and why.
New Rochelle, New York · Building since 1998
OPEN CODE✳LOCAL INTELLIGENCE✳REAL-WORLD SYSTEMS✳LEAVE A TRAIL✳
01 / Contributing upstream
Good code travels.
Recent contributions merged into Apple’s MLX and MLX-LM projects. Each links to the patch and review.
Apple MLXMerged
Repair quantized loader indexing
Corrected scale and bias indexing for small groups, with upstream dispatch changes in the merged patch.
#4202Aug 12, 2026Apple MLX-LMMerged
Keep converted Qwen weights correct
Prevented a second normalization shift when converted checkpoints retained MTP tensors.
#1623Aug 18, 202602 / Selected work
Less talk. More built.
Production systems, hands-on research, and the open resources behind them.
Open source
Wisp Coder
A code model trained from scratch to explore fill-in-the-middle and native multi-token prediction on Apple Silicon.
Enterprise AI
Enterprise agent platform
MCP integrations, reusable skills, and approval workflows connecting AI to the systems a business already uses.
Open source
Local inference on the Mac
MLX conversions, native MTP work in MTPLX, and SSD-backed expert streaming with iliria.
Enterprise AI
Retrieval with release checks
RAG systems with evaluation across retrieval, answer grounding, latency, cost, and regression behavior.
The whole Apple Silicon AI collection
Models, datasets, live demos, and technical guides — connected in one place.
Philip John Basile03 / The person behind the patches
Still building.
Still questioning.
I’ve spent my career turning complicated systems into things people can use.
That has meant enterprise platforms, data infrastructure, and now AI agents and local inference. I care about the details that survive a demo: correctness, clear boundaries, and a path someone else can follow.
More about my background →04 / Notes from the work
Leave a paper trail.
Essay
MCP is a governance problem ↗
What changes when a model can act on the tools and data inside a company.
Free field guideLearn by building the loop ↗
Eight practical missions covering agents, prompting, retrieval, evaluation, and tool use.
Research recordWhat the Wisp experiments showed ↗
Methods, comparisons, null results, and the limits of a small code model.