ashleygierisch/ESP32-AI
About the project
Running a 28.9M-parameter language model directly on an $8 ESP32-S3 microcontroller: text is generated on-chip with no server connection, at about 9.5 tokens per second.
Useful for
- Build a standalone device that generates short stories on a screen without internet
- Study the Per-Layer Embeddings method for fitting a large model into microcontroller flash
- Reproduce training and quantization on TinyStories using the code in src/ and experiments/
README summarized by DeepSeek V4.1 Flash. Details may be inaccurate.
Why it’s trending
- The repository is 5 days old and already has 61 stars.
- Top new repositories this week: #177.
Stars per day
Bars are daily stars, the line is the usual pace. Red marks spike days.
Numbers
- Total stars
- 61
- Stars in a day
- 0
- Forks
- 0
- Issues and pull requests
- 0
- Watchers
- 0
- Language
- Python
- License
- MIT
- Created
- September 9, 2026
- Last push
- September 9, 2026
Star trust
Growth looks organic: forks and discussion are in line with active projects, and stars arrive unevenly, the way people give them.
These are heuristics, not a verdict: we judge by the repository’s behavior, not by a list of stargazers.
Spotted in
- September 14, 2026Top new repositories this week: #175
Share
README badge
Paste it into your README — the badge shows the current magnitude and links to this page.
[](https://gitnova.dev/en/r/ashleygierisch/ESP32-AI)Similar projects
-
4.8
agamrossen/VolAnti
Open-source acoustic drone detection: listens for propeller harmonics, not radio, so it works against fibre-optic FPV drones that emit no signal. ESP32-S3 device with four microphones, e-paper screen and LoRa for alerts.
-
3.5
jomjol/AI-on-the-edge-device
Turns analog utility meters (water, gas, electricity) into digital ones using an ESP32-CAM: the camera photographs the readings and TensorFlow Lite recognizes the digits on-device. Data is exposed via MQTT, InfluxDB or…
-
3.4
hanyang9/UMR
Official codebase of UMR, a motion retargeting method that transfers human motion to humanoid robots via learned point cloud surface correspondence. It adapts motions from various sources (SMPL-X, BVH, GRAIL, OMOMO,…
-
3.0
amagine-ai/Amagine3D
An open-source 3D design layer for hardware creation: given a product description, reference images and dimensions, it generates editable enclosures and assemblies with STEP, STL and 3MF export.
-
2.8
Atmosphere-NX/Atmosphere
A custom firmware (CFW) for the Nintendo Switch that replaces the bootloader, TrustZone, sysmodules and patches Horizon OS to enable homebrew and extended features.
-
2.7
unitreerobotics/unifolm-wla
Unitree Robotics' 6B-parameter foundation model for humanoid robots, trained on ~2,500 hours of real-robot data and coordinating 64 manipulation tasks. Some weights and datasets are already open-sourced.