ri-esp-llm 0.1.1

no_std embedded inference primitives: int4 packing, KV cache, RNN helpers, sampling, compressed attention
Documentation

ri-esp-llm

Embedded inference primitives.

ri-esp-llm architecture

What it provides

int4 packing, KV cache, RNN helpers, sampling, compressed attention.

It is dependency-light and designed for embedded reuse.

Install

[dependencies]
ri-esp-llm = "0.1"

Quick start

// See crate docs/tests for complete examples.

API shape

  • no_std by default
  • fixed-capacity data structures where strings/payloads are needed
  • explicit receipt/schema constants where the crate crosses firmware/host boundaries
  • small public API intended to be readable in firmware reviews

Verification

From the workspace root:

cargo test -p ri-esp-llm
cargo +esp check -p ri-esp-llm --target xtensa-esp32s3-none-elf -Z build-std=core,alloc

This crate is also covered by the workspace gate:

cargo test --workspace

Integration path

Use this crate as one boundary in the larger ESP32 physical-AI stack:

sensor reading -> deterministic policy -> local-language receipt -> proof receipt -> display/log/optional escalation

Claim boundary

This crate is a reusable building block. It does not certify physical wiring, safety-critical behavior, production deployment, or model quality outside tests/receipts stated in the repository.

License

MIT OR Apache-2.0