After user feedback and further optimization, mlx-geospatial 0.1.1 is out! Faster spatial joins + kNN on Apple silicon. Warm-index component benchmarks: M1 Ultra: 18–1,262× (median 95×) M4 Max: 21–2,005× (median 105×) CPU/MLX ratios, shown on the same scale. 1/4

metal-graph 0.1.1 is out! ⚡ Millions of links. Hub and authority scores in milliseconds. This update brings HITS graph ranking to Apple Silicon GPUs through native Metal kernels, with Python and C APIs. Here are the M4 Max results. 🧵 1/7

Introducing metal-graph 0.1.0, a new Python library for fast graph analytics on Apple Silicon Macs. metal-graph combines a familiar Python interface with optimized native CPU code and GPU acceleration through Apple’s Metal framework. Its goal is simple: make demanding graph workloads faster while keeping computation local and fitting naturally into existing Python workflows. The first public release includes immutable graph structures, external node ID support, optimized CPU and Metal execution paths, and packaging built specifically for modern arm64 Macs. Getting started takes one command: pip install metal-graph This first release includes full-vector PageRank; batched personalized PageRank with top-k retrieval; multi-source, direction-optimizing breadth-first search with dense or sparse output; bounded k-hop neighborhood and ego-subgraph extraction; and experimental weakly connected components. The early results are exciting. On an M4 Max, metal-graph ran equivalent dense-output BFS on Orkut’s 117-million-edge graph in 8.29 milliseconds, compared with 41.2 seconds for rustworkx - a 4,965× speedup. In the included 1.18-million-edge quickstart benchmark, GPU PageRank was 109× faster than NetworkX, while the CPU path was nearly 60× faster. PageRank reached 614× versus igraph on Orkut, and weakly connected components reached 75.6× versus igraph on a 268-million-edge graph. Batched PPR with top-k achieved a contextual 1,059× speedup over a sequential igraph query loop. These are workload-specific, warm, end-to-end measurements from one M4 Max - not universal guarantees. The repository publishes the full methodology and clearly identifies comparisons where solver behavior or output representation differs. metal-graph supports CPython 3.10–3.14 on Apple Silicon Macs running macOS 14 or newer. Install it from PyPI or explore the benchmarks and notebooks in the GitHub repo (link below)

We are very excited to share one of the big projects that we have been working on at @Tabul_AI Metal TreeSHAP. Shapley Values are one of the most reliable and most widely used tools in Machine Learning interpretability. 1/11

I am excited to share with you a couple of projects that I've been working on at @TabulAI - Open Sync and Open Dashboard. These both started as side projects as I was looking for a solution that would help me work with minimal friction across my home network computer on my "main" projects. The solutions I've found in the marketplace were inadequate, and did not address some of my own main requirements. So I built my own and decided to open source them. Open Sync is a desktop SSH discovery, connection management, and port forwarding. GitHub repo: github.com/tabulai/open-sync Open Dashboard is local React and Express app for connecting to a remote SSH host, starting JupyterLab or classic Jupyter Notebook there, and opening it locally through an SSH tunnel. The UI also polls remote system memory and optional GPU utilization so the user can see basic host health next to the tunnel controls.. GitHub repo: github.com/tabulai/open-dash…