Showcase / question: a board-proven offline language runtime on ESP32-C3, and how you’d think about this relative to tiny dense LMs on microcontrollers

Hi llama4micro folks,

I wanted to share a small but unusual language-runtime project that feels closely related to the broader question your repo explores: what happens when language capability is pushed all the way down to microcontroller-scale hardware?

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 tiny dense language model running directly as a familiar inference loop. It is closer to a task-specialized language runtime whose
behavior has been crystallized into a compact executable form under severe physical constraints.

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

Why I’m posting here is that llama4micro is one of the clearest public examples of “language models on microcontrollers” as a real engineering path.

What I’d be curious about is whether systems like this should be thought of as:

• a separate family from tiny dense LMs on MCU
• an extreme endpoint where some task capability is better deployed as a dedicated executable runtime
• or part of a broader spectrum where language capability on tiny hardware may split into both small dense models and highly specialized runtime forms

Would be very interested in your thoughts.

No repository file, test, or entry point is named. Start by reviewing this issue and the linked Engram repository; a useful outcome would require a maintainer decision on whether the showcase belongs in llama4micro and a concrete scope, since the post currently defines no code change or completion criteria.