MLX

oMLX is an open-source LLM inference server for Apple Silicon Macs, managed through a macOS menu-bar application. It serves text, vision, OCR, embedding, and reranking models through compatible API endpoints, with model discovery and management, monitoring, continuous batching, and caching. Its caching system keeps KV cache in a hot in-memory tier and a cold SSD tier, allowing cached context to remain reusable across requests even when the conversation context changes. The project can be installed as a macOS app, through Homebrew, or from source, and includes a CLI for controlling the app-managed server. The repository states that it requires macOS 15 or later, Python 3.11–3.13, and Apple Silicon hardware.

MTPLX is an open-source native Mac app and command-line tool for running local language models on Apple Silicon. It uses a model's built-in multi-token prediction (MTP) heads to draft several tokens, verifies the block in one batched forward pass, and commits tokens with exact rejection sampling and residual correction, without requiring a separate draft model or changing the model's sampling distribution. It provides local OpenAI-compatible and Anthropic-compatible APIs, native chat, model management, hardware-specific draft-depth tuning, and tooling for building and verifying MTP models. The server can also expose MLX embedding and reranking models. MTPLX requires an M1-or-newer Mac running macOS 14 or later, and the repository is licensed under Apache-2.0.

kimi-k3-mlx is an MLX port for Apple Silicon of Moonshot’s Kimi K3 native-multimodal mixture-of-experts model. It provides a stock `mlx-lm`-compatible text model definition, a 3D video-capable MoonViT vision tower with a patch-merger projector, and an `mlx-vlm`-shaped wrapper that joins the language and vision components. The model has a 1-million-token context and uses routed and shared experts, Kimi Delta Attention, gated multi-head latent attention, Attention Residuals, latent-dimension experts, and the SiTU-GLU activation; routed experts are distributed in MXFP4 while other weights use bf16. The multimodal processor expands each image placeholder into a feature block rather than performing a same-length one-token-per-image scatter. The repository includes a streaming converter because standard `mlx-lm` conversion would materialize the multi-terabyte bf16 model, along with REAP expert-pruning scripts and expert-overlap analysis using Chinese, English, code, or mixed calibration data. It also contains vision-parity tests against Moonshot’s reference implementation. The repository states that the full model cannot run on a single Mac: even its smallest deployment tier requires hundreds of gigabytes of memory and exceeds the capacity of the largest Apple Silicon machine.

Laya-CoreML is an open-weight on-device decision-model runtime and port of Laya for Apple Silicon, using Core ML and Apple's Neural Engine. It answers typed choice, ordinal-score, and boolean questions with probabilities or scores rather than generating text or JSON; its Python API loads a model and returns structured answers. The repository includes Core ML checkpoints, tokenizer and configuration files, an offline Python package, a command-line Snake demo, and conversion and benchmarking tools. The Neural Engine bundle uses a short fixed-capacity input path, while longer or general-purpose requests can use CPU-and-GPU checkpoints. The project documents a separate Neural Engine graph using BC1L activations, 1×1 projections, and per-head attention, with CPU handling input and output boundaries. It supports macOS 15 or later on Apple Silicon and can run without PyTorch, Transformers, or MLX at inference time. The project is licensed under Apache-2.0 and describes itself as an independent port by Convai Innovations and contributors, not an official Convai Innovations or Apple release.

edge0 is an open-source streaming Mixture-of-Experts inference framework for running sparse MoE models on Apple Silicon Macs. Its MLX backend memory-maps expert weights from SSD storage and loads them on demand, so peak memory depends on the active experts rather than the model's total parameter count. A trained prerouter predicts the next expert routes so storage reads can overlap the forward pass, while Recover-LoRA adapters distill from the full-precision model to compensate for quantization loss; the base weights remain read-only and the adapters stay separate. The framework ships two end-to-end model tiers with bundled 4-bit checkpoints, LoRA adapters, and prerouter heads. It provides a command-line interface and Python API, supports demo, chat, and server modes, and exposes an OpenAI-compatible `/v1/chat/completions` endpoint. The current backend supports macOS on Apple Silicon M1 through M4; CUDA support is identified as planned. The repository is licensed under Apache-2.0.