New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
katago: init at 1.3.5 #82082
katago: init at 1.3.5 #82082
Conversation
The specific versions of cudnn and cudatoolkit is because they have to match, CUDA performance is actually quite bad reportedly for 9.2 and KataGo, so I selected 10.0, which is the version that the author mentioned he was originally running when he wrote this program. This program also is tested as working on cuda 10.2, but it requires using a version of cudnn that has not been accepted into the nixpkgs repository yet. |
f192e93
to
82f5b82
Compare
@GrahamcOfBorg build katago katagoWithCuda Please and thank you. |
82f5b82
to
d837724
Compare
Updated default CUDA Toolkit from 10.0 to 10.1 because that's what the repository readme suggested. |
d837724
to
b7d6b4b
Compare
@rasendubi @erikarvstedt @averelld Hi there, you were active on the original pull request for Leela Zero. If it's not too much trouble, can one of you review this pull request? |
Great that this is finally being packaged! I think just defaulting to tcmalloc would be better if we even make it configurable. Also, the buildbot probably won't evaluate anything with cuda because of it's unfree nature. Other than the above, LGTM, the quicker we merge the better! |
919b994
to
e6c9040
Compare
Ugh, I need to be more careful about screwing around with files locally when they still might change later. This should work now. @GrahamcOfBorg build katago katagoWithCuda |
8ab5936
to
67f8f4d
Compare
New release, packaged that instead of the old release. EDIT: Author updated the tag, so the latest force push is updating the sha commit to match the new tag. |
This pull request has been mentioned on NixOS Discourse. There might be relevant details there: https://discourse.nixos.org/t/prs-ready-for-review-may-2019/3032/137 |
67f8f4d
to
e901f5c
Compare
@GrahamcOfBorg build katago katagoWithCuda Once again into the night we go. |
You may have to restrict the platforms: it seems to require SSE, which rules out i686 and aarch64. |
No, I don't think there is anything more specific. |
e901f5c
to
e21da63
Compare
e21da63
to
c703763
Compare
@GrahamcOfBorg build katago katagoWithCuda |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
The expression looks good to me and ofBorg builds it.
I don't have the hardware to try, though, maybe only openCL.
I'd wait a bit to give @averelld the possibility to finish the review.
I just rebuilt with v1.3.4, with and without cuda still works great. |
c703763
to
dc3a76a
Compare
He released another minor version to fix a few newly discovered bugs. Sorry about the second version jump in this init pull request. |
It's a super short build, I like being up to date (and still works for me). |
All right, let's merge this. Thank you @OmnipotentEntity! |
Motivation for this change
Katago is the current strongest open source go AI engine. This derivation provides both the OpenCL (default) and CUDA versions of the engine.
Things done
sandbox
innix.conf
on non-NixOS linux)nix-shell -p nixpkgs-review --run "nixpkgs-review wip"
./result/bin/
)nix path-info -S
before and after)