M5Stack LLM Module Kit AX630C

With the M5Stack Module LLM Kit, you bring powerful, locally running AI directly to your device – without cloud or internet connection. The set consists of the Module LLM for offline inference and the Module LLM Mate with versatile interfaces for system integration. At the core is the AX630C SoC with Dual-Core ARM Cortex-A53 and dedicated NPU reaching 3.2 TOPS at INT8 – ideal for voice assistants, smart controls, or edge-based AI applications. With 4 GB RAM (3 GB of which for the NPU) and 32 GB eMMC storage, you can run multiple models simultaneously, including text, speech, and image processing. Pre-installed is the Qwen2.5-0.5B language model with wake-word detection (KWS), automatic speech recognition (ASR), language generation (LLM), and text-to-speech (TTS). The kit supports Arduino and UiFlow as well as the in-house StackFlow framework and can be configured and debugged via USB, UART, or network. Power supply is provided through the M5BUS interface, allowing the kit to seamlessly integrate into existing M5Stack environments. Particularly convenient: you can easily perform firmware updates via SD card or Type-C port. For demanding AI projects in education, industry, or private applications, this kit is just right.

Features:

• The core is the AX630C SoC with 3.2 TOPS inference performance at INT8, specially optimized for transformer models.
• With 4 GB LPDDR4 RAM (3 GB for hardware acceleration) and 32 GB eMMC storage, the module offers high performance for parallel model processing.
• The powerful Qwen2.5-0.5B model with wake-word, speech recognition, language generation, and TTS function is pre-installed.
• The Mate module complements the functionality with USB-to-Serial (CH340N), Ethernet (RJ45), and M5BUS power supply.
• Thanks to the SD card slot, USB-C port, UART, and RGB status LEDs, development, debugging, and maintenance are efficiently possible.
• The entire system operates with only about 1.5 W under full load – ideal for continuous operation.
• The kit is plug-and-play capable and compatible with other M5Stack platforms.
• Tutorials and libraries for Arduino, UiFlow, and StackFlow ease startup for you.

Application Examples:

• Voice assistant in offline mode
• Text-to-speech converter in embedded systems
• Control of smart home devices without the cloud
• Interactive robots with local speech understanding
• Expansion of industrial equipment with AI functionality

Documentation:

• M5Stack Wiki
• Official M5Stack Forum

Technical Specifications:

• Manufacturer: M5Stack
• Manufacturer SKU: K144
• Processor: AX630C Dual Cortex-A53 @ 1.2 GHz with NPU (3.2 TOPS @ INT8, 12.8 TOPS @ INT4)
• Memory: 4 GB LPDDR4 (1 GB System, 3 GB NPU), 32 GB eMMC5.1
• Audio: Microphone MSM421A, Audio Amplifier AW8737, Speaker 8 Ω @ 1 W
• Communication: USB-Serial (CH340N), RJ45 Ethernet, UART, FPC-8P, SD card upgrade
• Features: KWS, ASR, LLM, TTS, RGB LED Status Display, Ubuntu System
• Power Consumption: 5 V @ 0.5 W (Idle), 5 V @ 1.5 W (Full Load)
• Dimension: 54x54x13mm
• Weight:  66g

Package Contents:

• 1 x Module LLM
• 1 x Module LLM Mate
• 2 x FPC-8P Cable