Question: does a flash-resident tiny language runtime fit the MCUNet view of tiny system-algorithm co-design? #39

Hi MCUNet folks,

I wanted to share a small MCU language-runtime experiment and ask whether systems like this feel adjacent to the MCUNet view of tiny AI systems.

We built a public demo line called Engram and deployed it on a commodity ESP32-C3.

Current public numbers:

• Host-side benchmark capability

  • LogiQA = 0.392523
  • IFEval = 0.780037

• Published board proof

  • LogiQA 642 = 249 / 642 = 0.3878504672897196
  • host_full_match = 642 / 642
  • runtime artifact size = 1,380,771 bytes

Important scope note:

This is not presented as unrestricted open-input native LLM generation on MCU.

The board-side path is closer to a flash-resident, table-driven runtime with:

• packed token weights
• hashed lookup structures
• fixed compiled probe batches
• streaming fold / checksum style execution over precompiled structures

So this is not a standard tiny dense model with familiar operators. It is closer to a task-specialized language runtime whose behavior has been
pushed into a very compact executable form.

Repo:
https://github.com/Alpha-Guardian/Engram

What makes me curious from the MCUNet perspective is whether systems like this should be viewed as:

• outside the normal tiny-DNN scope
• an adjacent form of system-algorithm co-design for language workloads
• or evidence that some language-task behavior can also be compressed into MCU-scale specialized runtimes

Would be very interested in your thoughts.