I need to start up the Patreon page early in attempt to get funding for more RAM since my computer is crashing every day due to a high ram usage EOF Error.
Significance of this issue:
I am unable to train BetaOne continuously due the daily crashes and as a result, this will likely put back training significantly if not fixed.
Pre-Build BetaOne support rewards:
For helping Pre-build BetaOne, you will receive an honorary tag/role of Pre-BetaOne supporter. This is only for a limited time.
What is BetaOne?
BetaOne is a neural network that uses the non-zero approach to produce the strongest chess entity possible.
How Strong is BetaOne? (Training still in progress)
An official release can be found in the posts above.
Why support BetaOne?
Due to hardware limitations, I had to compromise strength for speed. One example would be the network architecture.
A 40b run for example would take me 70 days on my hardware. This unfortunately is just not feasible for multiple runs.
Being a full time college student, I cannot afford to upgrade my hardware. With help from supporters, I would be able to purchase new hardware that I can use to develop BetaOne further.
BetaOne is and will always be a free neural network for the community. The focus of this project is not only to push the limits of chess, but to provide entertainment to all in computer chess tournaments.
What will funding be used for?
Hardware upgrades to allow for faster training runs.
With better hardware, I will be able to get networks out within 2-3 weeks instead of the months it currently takes.
RAM/SSD upgrades to solve the bottlenecking issue.
16 GBs of ram significantly slows down training in which weeks are lost overall.
Currently my SSD is unable to hold the 330 GBs of training data. I must split the files up into 1/6 of total (88M files, 25M Games)
This does not include storing the networks, checkpoint training files, logs, Windows 10, or anything else. When factoring in backups, around 1 TB is used.
With funding, I will be able to purchase better hardware to resolve these bottlenecking issues with RAM and SSD space.
If enough funding can be secured per month, GPUs can be upgraded to speed up training drastically. (2060 FP16 Training)
With funding, I will be able to focus more time on BetaOne development.
So far I have placed hundreds of work hours and thousands of computational hours into the project.
By supporting this project, I will be able to focus more on BetaOne development and produce stronger networks for the community.
FAQ and Links:
For more technical information please visit the BetaOne/Stockfish discord server here: