AI & ML interests
None defined yet.
Recent Activity
π Welcome to the AIFoundry CORE-ET Hackathon
Port real models. Run them on real silicon. Share the results. β‘
This is a community hackathon for people who want to make open AI workloads run closer to the edge: vision, speech, audio, local language models, robotics, drones, and weird benchmark ideas that deserve to touch hardware. π€
The hackathon is already moving: 300+ participants, 20+ submissions, community board runs, and real performance improvements in the first week.
π $10,000 in HF credits + 5 hardware boards
In partnership with Hugging Face, OpenHW Foundation, and AI Foundry, we are giving participants and winners:
- 500 Γ $20 in Hugging Face Inference credits, sponsored by Hugging Face.
- 3 Digilent Arty A7-100T FPGA boards, sponsored by OpenHW Foundation.
- 2 ET-SoC1 cards, sponsored by AI Foundry.
π€ HF Inference credits for the first 500 participants
There is no hard challenge to solve before claiming the credits. Join the
Hugging Face hackathon organization,
join our Discord, and DM
core-et hk bot#3838 with your Hugging Face handle.
The credits are enough to start experimenting with open models while you work toward a model port, performance result, or hardware prize.
β‘ Competition tracks
| Track | How to win | Prize | Deadline |
|---|---|---|---|
| Weekly Challenge 1: YOLO on CORE-ET | Record the fastest valid official YOLO run on ET-SoC1. Lowest kernel_wait_s wins. |
1 Digilent Arty A7-100T | July 11, EOD AoE |
| Weekly Challenge 2 | A new focused hardware target and metric will be announced next week. | 1 Digilent Arty A7-100T | July 17, EOD AoE |
| Best Llama 3.2 1B performance | Record the highest valid decode tokens/second result on ET-SoC1. | 1 ET-SoC1 card | July 24, EOD AoE |
| Most approved model ports | Finish the hackathon with the highest number of reviewed and approved model ports. | 1 ET-SoC1 card | July 24, EOD AoE |
| Community Prize | Help others through docs, recipes, debugging, board access, and reproducible runs. | 1 Digilent Arty A7-100T | July 24, EOD AoE |
One hardware board maximum per participant or team. A valid entry must leave enough proof for another builder to inspect and reproduce the result: model, board, metric, proof.
π Ready to join?
First step: join the Hugging Face organization so you show up with the rest of the teams, follow shared resources, and get the latest hackathon updates.
π Join the AIFoundry Hackathon Org
Bring your teammates too. The org page is the public roster for people, teams, models, demos, and board results around the hackathon. π
π§© Pick a quest
| Quest | What you can build |
|---|---|
| ποΈ Vision | Denoising, detection, camera-first edge workloads |
| ποΈ Speech + audio | Whisper-style kernels and streaming-friendly pipelines |
| π§ Local LLMs | GGUF and llama.cpp flows on ET-backed runtimes |
| π€ Robotics + drones | Perception, control, and low-power deployment paths |
| βοΈ Systems | Faster kernels, better memory movement, cleaner board runners |
| π§ͺ Wildcard | Bring your own open model and make the benchmark reproducible |
π¦ Model repo
The model repo is the technical home for examples, docs, benchmark artifacts, and the submission path:
π Open the CORE-ET model repo
Start from an existing port, improve a kernel, add a new model, or package a small demo that shows why your port matters.
π§ How participation works
- π€ Join this Hugging Face organization.
- π― Team up or go solo.
- π― Pick a model, kernel, benchmark, or demo target.
- π Use the model repo to follow the technical path.
- π Submit your port or improvement through a GitHub PR.
- π The board workflow runs and comments back with results.
π± Good first contributions
- β‘ Improve an existing DnCNN, YOLO, Whisper, or LLM benchmark.
- π Add a new open model with a pinned Hugging Face reference.
- π Make a kernel faster and document the board result.
- π¬ Package a small demo that shows the workload in action.
- π«Ά Help another participant reproduce a run.
π¬ Need help?
Join Discord and ask in #community-lab:
Tell us what you want to port, what hardware access you need, and whether you are looking for teammates.