Hack on ARC Prize or Fine-tune Deep Seek: Strong Compute + GPUs included
Previous winners and grantees:
Deep Seek Fine Tuning: Hackathon: Increase PyTorch -> CUDA translation from 10%->30% accuracy
ARC Prize: Grantee placed Second Place in ARC Prize contest
Chess Bots: Train From Scratch: reached 2000 ELO in 10 hours.
STRONG COMPUTE GPU HACKATHON
For Engineers, AI Researchers, students & anyone who is comfortable with PyTorch.
Our goal is to offer the best platform to build and train models at cluster scale. Come on over, light up some clusters, have a blast and let us know what you love and what needs work.
Strong Compute’s Instant Super Computer (ISC) is designed to be the easiest way to get started on multinode training. You’ll be up and running in an hour.
Engineers only. All code. Recruiters/Slidegineers/Marketers etc. need not apply.
All applicants will be vetted for technical fit before being approved to attend.
Choose from one of the following hacks to compete in over the weekend:
Competition A: ARC Prize Challenge
Winner will earn compute on Strong Compute through our research program to work on the 2025 ARC Prize*
Join a curated, interview-only hackathon alongside top AI researchers, engineers, and developers to tackle the ARC Prize.
The ARC Prize is a $750,000+ public competition to beat and open source a solution to the ARC-AGI-2 benchmark—learn more here. ARC-AGI-2, launched in March 2025, sets a new standard for AI reasoning: every task is solvable by humans in under two tries, yet unsolved by any frontier AI system.
Last year, we supported a project that went on to place 2nd in the 2024 ARC Prize. This year, Strong Compute is proud to return as a Foundation Partner — backing more open-source efforts pushing the frontier of machine reasoning.
Join us over a weekend at our hacker house. We’ll bring the GPUs, food, drinks, desks, and vibes — you bring your laptop and your ideas.
Details:
Challenge: Build a research-driven, technically promising solution to the ARC Prize challenge.
Resources: Past ARC benchmark tasks, evaluation methods, open research papers, and dev environment.
Compute: Dedicated GPU access via Strong Compute’s ISC.
Judging: Evaluation of research rigor, novelty, and model performance against ARC-AGI benchmark.
Competition B: Deep Seek Fine-Tuning
DeepSeek has landed. Come build the next demonstration of leaner, meaner reasoning models dominating the big guys.
Strong Compute will provide the DeepSeek distillation model weights at your fingertips, and a bunch of GPUs to fine-tune on. Over to you!
Fine-tune a DeepSeek reasoning model on your dataset and show us what it can do. The most impressive demo wins.
Details:
Challenge: Fine-tune a DeepSeek model on your dataset.
Resources: Pre-downloaded DeepSeek model variants and a demonstration of how to fine-tune on Strong Compute.
DeepSeek-R1-Distill-Qwen-1.5B
DeepSeek-R1-Distill-Qwen-7B
DeepSeek-R1-Distill-Llama-8B
DeepSeek-R1-Distill-Qwen-14B
DeepSeek-R1-Distill-Qwen-32B
DeepSeek-R1-Distill-Llama-70B
Compute: Dedicated GPU access via Strong Compute’s ISC.
Judging: Qualitative assessment of what your fine-tuned model can do that the stock model can’t.
💰 Prize
$2.5K–$25K Open Source Research Compute Grant* for each competition. 1 Prize for ARC and 1 Prize for Deep Seek.
📍 Location
San Francisco, USA – Venue details will be shared upon successful application.
Address: West Portal, SF
🗓️ Event Schedule
Friday 17 April 2025
6:00 pm: Arrive. Drinks, snacks + networking
7:00 pm: Set-up, onboarding, test training run + inference
7:30 pm: Competition begins
8:30 pm: Dinner
Late: Hacking
Saturday 18 April 2025
9:00 am: Day two start, breakfast
1:00 pm: Lunch
4:30 pm: Draft submissions
5:45 pm: Final submissions
6:00 pm: Dinner + Presentations
8:30 pm: Wrap up
Strong Compute
Strong Compute provides infrastructure management capability for rapid artificial intelligence development.
P.S. We’re hiring. This is a great way for us to get to know each other. Open roles: https://jobs.lever.co/strongcompute
Community Partner:
Build Club – The home for top AI engineers in APAC.
*Winners must finalize an acceptable research proposal within 30 days after the event to be eligible for compute grants. Refer to our research grant requirements.