sipeed/TinyMaix

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1071 stars · 169 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

TinyMaix is a tiny neural network inference library for microcontrollers (TinyML), with core code under 400 lines and a .text section under 3KB. It supports INT8/FP32/FP16 (and experimental FP8) models converted from Keras h5 or TFLite, with SIMD acceleration for ARM, RISC-V, CSKY, and x86.

Use cases

• run mnist inference on an Arduino ATmega328
• deploy neural network models on resource-limited microcontrollers
• run tinyml models on embedded chips with only 2KB of RAM
• convert keras or tflite models for mcu inference
• accelerate inference with ARM SIMD/NEON or RISC-V vector extensions
• train and deploy models online via MaixHub

When to choose

• you need neural network inference on very small MCUs with limited flash and RAM
• you want a simple, readable TinyML library you can understand in 30 minutes
• you need multi-architecture support including ARM, RISC-V, and x86
• you want full static memory configuration

When to avoid

• you need a feature-rich production TinyML stack with CMSIS-NN optimizations
• you need large or complex models beyond tinyml scale
• you need many advanced ops and new model architectures
• you want a mature ecosystem like TFLite Micro or microTVM

Facets

library · maturity active

machine-learning llm-inference embedded machine-learning embedded-systems artificial-intelligence embedded c cross-platform iot tinyml neural-network-inference microcontroller int8-quantization mcu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="sipeed/TinyMaix")