DeepLearningKit
Open-source (Apache 2.0) GPU-accelerated deep learning framework for Apple iOS, OS X, and tvOS, alongside caffemodel2json (33★).
Hands-on software, benchmarks, and open-source libraries across Mojo, Python/PyTorch, Swift/Metal, and C/C++ — from 2010s search & GPU engines to current Mojo SIMD libraries and agent tooling.
GPU deep learning frameworks, Apple Metal & CoreML compute, and PyTorch training experiments.
Open-source (Apache 2.0) GPU-accelerated deep learning framework for Apple iOS, OS X, and tvOS, alongside caffemodel2json (33★).
Curated community bibliography of GPU deep learning publications and early accelerator research.
Early iOS 8 Swift + Metal GPGPU benchmarks comparing Metal compute shaders, shared CPU/GPU memory, and Apple Accelerate.
End-to-end example for converting and running Keras deep learning models on iOS via CoreMLTools.
Metal GPU compute examples for macOS (29★) and Apple tvOS (memkite/MetalForTVOS, 9★).
Transformer grokking and Muon vs. AdamW optimizer dynamics accompanying arXiv:2504.16041.
Reducing deep learning and functional approximation to a single continuous Exp-minus-Log (EML) Sheffer primitive.
Distributed Hadoop/MapReduce algorithms, C/C++ storage & search engines, and analytics tooling.
Parallel machine learning and data-processing algorithms in Python for Hadoop MapReduce.
Large-memory key-value pair store implemented in C with Python bindings for high-throughput workloads.
Lightweight Python client library for Amplitude analytics event logging.
Machine-learning search ranking in C++ alongside an inverted-index search engine built on Amazon SimpleDB.
Recent pure-Mojo AI agent tooling, 6.5 GB/s SIMD parsing, tokenization, and enterprise libraries.
A pure Mojo port of the PI AI Agent Toolkit for fast, low-overhead agent workflows.
Pure Mojo RFC 8259 SIMD JSON parser hitting 6.5 GB/s on Apple M3 Ultra, outperforming Rust and C++ parsers.
Pure Mojo BPE, tiktoken, and HuggingFace-compatible tokenizer optimized for LLM inference on Apple Silicon.
26-module pure Mojo enterprise library collection spanning Redis, SQL, caching, circuit breakers, tracing, and testing.
Zero-copy inline vector buffers in Mojo engineered for predictable latency and cache-friendly SIMD access.