Releases·Tool Release
Dev builds Rust REPL/DSL and tiny MoE models running in-browser and on ARM/RISC-V boards to visualize MoE experts and Engram
An ML study-group member shared a browser-based 'MoE Microscope' built with a custom Rust REPL/array-programming DSL, demonstrating tiny dense and MoE models to visualize experts, routing, and Engram-style concepts. The tool targets browser (WASM), macOS/Linux CLI with Apple and NVIDIA GPUs, and eventually small ARM/RISC-V boards.
2 engagement1 sourceMon, Sep 14, 2026, 01:59 AM
AI Overview
- A Rust REPL/AOT-compilable array-programming DSL with ML concepts and visualization primitives runs in-browser via WASM and at CLI on Apple and Linux GPUs
- The dev built one dense and several tiny MoE model variants in a browser-runnable 'MoE Microscope' demo to experiment with Engram, KV cache compression, and agent optimizations
- Goals include tiny MoE models running on small ARM and RISC-V boards via no_std Rust, possibly using NPUs
- The author credits Freetoken with enabling larger MoE models to run on 12G–24G VRAM local GPUs
Related
Products·building
Visual ML programming language for learning and teaching ML concepts
A retired platform engineer built a visual ML programming language as an alternative to Python/Colab for learning and teaching ML. It runs training on CPU in browser, Apple Silicon via MLX, or NVIDIA via CUDA. The goal is to train local model coding agents to learn the user's style and process.
A retired platform engineer built a visual ML programming language as an alternative to Python/Colab for learning and teaching ML. It runs training on CPU in browser, Apple Silicon via MLX, or NVIDIA via CUDA. The goal is to train local model coding agents to learn the user's style and process.
0 engagement1 sourceredditMon, Aug 10, 2026, 02:01 AM
Products·building
sw-mlpl: open source array programming language playground for learning machine learning
An open source array programming language with a web playground for learning and visualizing machine learning concepts. The UI can run locally against Apple Silicon or NVIDIA GPUs, and can connect to Ollama models for LLM-based context-aware help in the REPL. Built by a developer after finding Google Colab inadequate for their ML study group.
An open source array programming language with a web playground for learning and visualizing machine learning concepts. The UI can run locally against Apple Silicon or NVIDIA GPUs, and can connect to Ollama models for LLM-based context-aware help in the REPL. Built by a developer after finding Google Colab inadequate for their ML study group.
3 engagement1 sourcehackernewsWed, Sep 2, 2026, 10:05 PM
Products·launched
Run massive MoE models on mobile devices via CPU
BigEdgeOnMoE is an open-source codebase that enables running large Mixture-of-Experts models (up to 120B parameters) on standard smartphones or consumer PCs using only CPU. It achieves 6 tokens/second for Qwen 35B (Q4) on a mid-range phone with 12GB RAM, making large models accessible without specialized hardware.
BigEdgeOnMoE is an open-source codebase that enables running large Mixture-of-Experts models (up to 120B parameters) on standard smartphones or consumer PCs using only CPU. It achieves 6 tokens/second for Qwen 35B (Q4) on a mid-range phone with 12GB RAM, making large models accessible without specialized hardware.
20 engagement1 sourceredditThu, Jul 30, 2026, 03:34 PM
Products·building
Vyact: open-source desktop workspace for local LLMs, documents, browser context, and coding
Vyact is a free, open-source desktop app that combines local model management, document RAG, browser context, and coding workflows in one place. It handles downloading and running local models (using MLX on Apple Silicon) so users can work with their own files alongside a local LLM. It targets people who want everyday AI tasks without juggling separate tools.
Vyact is a free, open-source desktop app that combines local model management, document RAG, browser context, and coding workflows in one place. It handles downloading and running local models (using MLX on Apple Silicon) so users can work with their own files alongside a local LLM. It targets people who want everyday AI tasks without juggling separate tools.
3 engagement1 sourceredditFri, Sep 4, 2026, 03:12 PM
Releases·Tool Release
runNburn: Rust GGUF runtime targets models larger than RAM
A developer announced runNburn, an open-source Rust inference runtime for quantized GGUF models that runs on CPU, CUDA, Metal, and Android, with experimental Vulkan and OpenCL paths. The runtime is designed to handle models that exceed available fast memory, addressing a gap in existing solutions.
A developer announced runNburn, an open-source Rust inference runtime for quantized GGUF models that runs on CPU, CUDA, Metal, and Android, with experimental Vulkan and OpenCL paths. The runtime is designed to handle models that exceed available fast memory, addressing a gap in existing solutions.
0 engagement1 sourceredditFri, Jul 31, 2026, 03:42 AM