โ๏ธ๐บ๐ธ Mega Chess SF Hackathon: Train AI Chess Bots on GPUs
โSpend 2 days training AI Chess bots on GPUs in SF or Sydney.
โ๐ TLDR:
โJoin our Mega Chess Hackathon to get hands-on experience with rapid iteration cluster training using Strong Computeโs Instant SuperComputer (ISC). This event will focus on training deep neural networks to play chess.
โUnlike traditional training environments, the ISC simplifies and accelerates job execution on shared GPU clusters, allowing training to start within seconds. Shared resources are efficiently managed, and large datasets are readily accessible. Experiment with different models, datasets, and hyperparameters to develop the best chess AI over two days.
โThe winner will receive a $10k-$100k compute grant for open-source AI research purposes*. ๐๐ฐ
โ
โ๐ Details:
โWeโre providing 2-day access to 48 GPU clusters. Twenty teams will compete, blending human intuition, tonnes of data, and innovative techniques to create the ultimate chess AI.
โThe full set of rules and gameplay will be shared with participants ahead of time on Discord.
โ
โ๐งฉ Task:
โTrain a deep neural network to play chess. Inputs to the neural network during inference (game play) will be (1) the current game history in Portable Game Notation (PGN) using Standard Algebraic Notation (SAN) and (2) a possible move in SAN. Inference output from the model will be a numeric score representing your modelโs strength of preference for that move.
โ
โ๐ Dataset:
โDatasets that will be made available include (1) board states evaluated by StockFish and (2) PGNs of historic games. Code for generating training data, as well as example data loading pipelines will be provided for reference. Participants can also use their own resources.
โ
โ๐ง Models:
โParticipants are free to implement any model architecture that satisfies the input and output constraints described above. Example model architectures will be supplied, which demonstrate application of computer vision (e.g. Convolutional Neural Network & Transformer) and autoregressive language model (e.g. Transformer) architectures. Models must meet memory and inference time constraints to be determined. Models must be trained from scratch during the competition.
โ
โ๐ป Compute:
โAccess to multiple 48x 24GB Ampere clusters and dedicated GPU workstations on the Strong Compute Instant Super Computer platform. Compute credits will be provided during the competition.
โ
โ๐ Judging:
โEach teamโs model will play off in a tournament to determine the competition winner.
โ
โ๐ Prize:
โThe winning team will receive one of Strong Computeโs $10-100K grants for open-source AI research or further development of open chess models.
โ
โ๐ Participant Requirements:
โExperience with PyTorch and cluster training is preferred but not mandatory.
โSkills in large-scale clusters (64+ GPUs) are highly valued.
โTeams of 1 to 3 members are allowed. Include your team membersโ names when registering. Approved participants will receive a link to our Discord competition channel.
โ๐ Locations:
โSan Francisco, USA: Venue details will be shared upon successful application. ๐บ๐ธ
โSydney, Australia: Register here ๐ฆ๐บ
โ๐๏ธ Event Schedule:
โSan Francisco:
โFriday:
โ6:00 pm: Drinks, snacks + networking
โ6:30 pm: Set-up, onboarding, test training run + inference
โ7:30 pm: Competition begins ๐
โ8:30 pm: Dinner
โLate: hacking
โSaturday:
โ9:00 am: Day two start, breakfast
โ1:00 pm: Lunch
โ7:30 pm: Dinner
โ8:00 pm: Final submissions ๐
โ9:00 pm: Competition begins โ๏ธ
โ11:00 pm: Winners announced ๐
โStrong Compute
โStrong Compute provides infrastructure management capability for rapid artificial intelligence development.
โP.S. We're hiring and this is a great way for us to get to know each other https://strongcompute.com/jobs
โSupported by community partner:
Build Club: the home for top AI engineers in APAC
โ*Refer to our research grants for requirements.