NixOS CUDA | 290 Members | |
| CUDA packages maintenance and support in nixpkgs | https://github.com/orgs/NixOS/projects/27/ | https://nixos.org/manual/nixpkgs/unstable/#cuda | 57 Servers |
| Sender | Message | Time |
|---|---|---|
| 7 Mar 2025 | ||
Wait a minute, I am slightly confused as llama-cpp seems to actually have cuda support now that I rebuilt a couple of minutes ago. It just does not use my GPU when running inference even though it reports it as visible and usable. Maybe a configuration mistake on my side (although I am using the default NixOS service). I'll look into open-webui and onnxruntime now... | 13:27:31 | |
| Yes, onnxruntime does recompile, as well as python3.12-torch-2.5.1. I'm checkin the hashes now... | 13:33:36 | |
| I am definitely building onnxruntime myself even though I get:
Which is the same hash as the hydra build store path | 13:48:48 | |
| I get the same hash for pytorch locally and in hydra as well! | 13:56:11 | |
And if you nix build --override-input nixpkgs github:NixOS/nixpkgs/9f41a78ead0fbe2197cd4c09b5628060456cd6e3 .#nixosConfigurations.$(hostname).pkgs.onnxruntime? | 13:59:42 | |
| Then I'm building nccl and cudnn-frontend for some reason? | 14:15:15 | |
| Well this certainly shouldn't be happening if the hashes indeed match | 14:21:40 | |
| Which hydra eval did you refer to? | 14:22:00 | |
| * Which hydra eval did you refer to, can you link it? | 14:22:06 | |
| Sorry, I am back now. It seems that my setup had complicated things: I was trying stuff on a laptop while the actual setup was on another host (with the actual GPU), but I did use the correct hostname for the workstation, which should (I mean, that is the whole point of Nix?) lead to the same build. (Both systems are x86_64) However I was also trying globally enable cudaSupport and package-overridden cudaSupport, which might have lead to me making a mistake, I don't know. All I can say is that
now just downloads onnxruntime from the cache, which is the expected behaviour. I'm going to check without overridden input and pytorch again and then with the whole system. | 15:27:49 | |
Building without the overridden nixpkgs input forces rebuild (I used just nix build .#nixosConfigurations.$(hostname).pkgs.onnxruntime) | 15:29:00 | |
| For earlier, I meant the derivation store path of https://hydra.nix-community.org/build/3297277#tabs-details | 15:39:13 | |
| Ok, python3.12-torch-2.5.1 is fetched from the community cache again iff I override the nixpkgs input again to the same hash as in https://hydra.nix-community.org/build/3534138#tabs-buildinputs | 15:43:38 | |
As in nix build --override-input nixpkgs github:NixOS/nixpkgs/e9b0ff70ddc61c42548501b0fafb86bb49cca858 .#nixosConfigurations.$(hostname).pkgs.python3Packages.pytorch | 15:43:55 | |
| If I don't, then it rebuilds | 15:44:23 | |
| Does that mean I have to find the right commit in nixpkgs that is somewhere between the onnxruntime hydra build, pytorch hydra build and my system that lets me fetch all of them? | 15:45:26 | |
| Oooorrr I just need to update my nixpkgs again. Weird, it was not that old, only a week or so. I guess that fixed it (apart from llama-cpp not actually using the GPU, but that is another issue). Thanks for the debugging help, I definitely learned new ways of debugging such problems here! | 15:52:49 | |
| 15:56:04 | ||
| Well, not so fast: I now get a build error of /nix/store/8msyislppkkr4dxzhir6qd2vc6qg23am-python3.12-rapidocr-onnxruntime-1.4.4.drv, which seems unrelated to https://hydra.nixos.org/build/291903538, which fails because of SDL's build failure (a dependency for some reason). I get a segfault somewhere in python in the pytestCheckPhase of the build. | 16:55:43 | |
That's why we use | 20:55:11 | |
| 8 Mar 2025 | ||
| @connorbaker:matrix.org i dont suppose your talk was recorded? | 12:23:13 | |
| It probably was, just taking them a while to upload. If not, the slides are available here: https://github.com/ConnorBaker/2025-planet-nix | 19:57:32 | |
| Unless I am misunderstanding something, looking at the Southern California Linux Expo youtube channel, it seems that the only nix related streams from the 6th and 7th of march are for Room 101. And the "Evaluating the Nix Evaluator" talk was scheduled for Room 2 (which I am assuming is actually Room 102)? I wonder... | 20:42:23 | |
In reply to @ruroruro:matrix.orgThere were two room tracks for PlanetNix | 23:57:15 | |
| It usually takes a month or so for the complete set of talks to be cut, edited and uploaded | 23:58:46 | |
| 9 Mar 2025 | ||
| oh TIL they have livestreams | 00:23:32 | |
| Room 2 for Planet Nix was conference room 104, which I didn’t see on the livestreams when I checked this morning 🤷♂️ | 02:25:53 | |
| 13:30:40 | ||
SomeoneSerge (UTC+U[-12,12]): sorry to bother you with this again, but could you please take a final look at https://github.com/NixOS/nixpkgs/pull/379768#discussion_r1977975372 and let me know if you prefer setting badPlatformsConditions/hydraPlatforms for broken packages or filtering them in release-cuda.nix? | 17:25:00 | |
| Also, apparently the "Evaluation Errors" tab missing for individual evaluations isn't a nix-community specific problem. The same thing happens with the official Hydra instance, so I opened an issue upstream: https://github.com/NixOS/hydra/issues/1453. | 17:28:52 | |