 | NixOS CUDA | 310 Members |
| CUDA packages maintenance and support in nixpkgs | https://github.com/orgs/NixOS/projects/27/ | https://nixos.org/manual/nixpkgs/unstable/#cuda | 60 Servers |
| NixOS CUDA | 310 Members |
| CUDA packages maintenance and support in nixpkgs | https://github.com/orgs/NixOS/projects/27/ | https://nixos.org/manual/nixpkgs/unstable/#cuda | 60 Servers |
| Sender | Message | Time |
|---|---|---|
| 10 Jun 2024 | ||
 connor (he/him) | Gaétan Lepage: did you mention there was a PR or something merged to disable the checkPhase or test suite for botorch, or did I misunderstand? | 14:01:56 |
| connor (he/him) | On another note, has anyone built elpa (https://github.com/NixOS/nixpkgs/blob/master/pkgs/development/libraries/elpa/default.nix) successfully with CUDA support? I let it run for like 20h and it was still building. Seems to compile four object files at a time? | 14:04:03 |
 Gaétan Lepage | In reply to @connorbaker:matrix.orgNo, I have not done anything. I was actually able to build it just fine from master earlier today. | 14:29:01 |
 hexa | Gaétan Lepage: have you considered pulling this patch for tensorflow-bin? https://github.com/tensorflow/tensorflow/issues/58073#issuecomment-2097055553 | 20:58:34 |
| 11 Jun 2024 | ||
 teto | when using localai 2.15 from unstable and even after a reboot I get ggml_cuda_init: failed to initialize CUDA: CUDA driver is a stub library. It's a bit random but if anyone has a tip, I take it. nvidia-smi output looks fine | 00:25:38 |
| Gaétan Lepage | In reply to @hexa:lossy.networkThis looks like it could work ! However, how do you apply a patch to a wheel-type python derivation ? | 06:38:47 |
| Gaétan Lepage | What phase of the buildPythonPackage script should I hook it into ? | 06:39:02 |
| Gaétan Lepage | I tried patches = [ but it does not work | 06:39:15 |
| Gaétan Lepage | I am packaging this: https://github.com/EricLBuehler/mistral.rs?tab=readme-ov-file#installation-and-build -> What should be the approach ? Adding | 07:01:41 |
 kaya 𖤐 changed their profile picture. | 08:03:48 | |
| hexa | In reply to @glepage:matrix.orglikely in postInstall 😕 | 11:58:18 |
| hexa | curses | 11:59:30 |
| Gaétan Lepage | Ok, but I can I use fetchpatch though ? | 12:02:43 |
 SomeoneSerge (matrix works sometimes) | connor (he/him) (UTC-5) IIRC you brought up setting legacy (FindCUDA&c) variables from the setup hooks. I think we should set them, and we should put that logic behind a guard (e.g. findCudaCmakeSupport=true), just as we should guard the current logic (e.g. findCudatoolkitCmakeSupport=true). We should disable the legacy by default. We should only set cmake flags when the cmake hook is actually used or when cmake flags are explicitly requested. | 13:19:22 |
| SomeoneSerge (matrix works sometimes) | In reply to @keiichi:matrix.orgLD_DEBUG=libs | 13:19:44 |
| SomeoneSerge (matrix works sometimes) | In reply to @glepage:matrix.orgDoes it allow enabling multiple features at once? | 13:20:13 |
| Gaétan Lepage | In reply to @ss:someonex.netNo, but I think that I will copy the implementation from ollama | 13:20:38 |
| Gaétan Lepage | It looks very clean to me | 13:20:44 |
| Gaétan Lepage | https://github.com/NixOS/nixpkgs/blob/master/pkgs/by-name/ol/ollama/package.nix#L65-L82 | 13:21:07 |
| SomeoneSerge (matrix works sometimes) | The shouldEnable logic looks maybe a bit complex but the arguments seem good? | 13:23:21 |
| SomeoneSerge (matrix works sometimes) | In reply to @gjvnq:matrix.orgLooking more closely, I'd guess the issue is somewhere around __has_include(<Imath/half.h>) in ${openimageio.dev}/include/OpenImageIO/half.h | 14:49:05 |
| teto | SomeoneSerge (UTC+3): TIL, LD_DEBUG looks quite useful. I suppose the "stub" referred to in the message concerns /nix/store/q3m473lh6gcg4xbhbknrhmcj7w7njjs6-cuda_cudart-12.2.140-lib/lib/stubs/glibc-hwcaps/x86-64-v3 . Do you know what a "stub" is and why that would be a problem ? I understand "stub" as a "generic" library ? (I have a 3060RTX) | 16:34:54 |
| connor (he/him) | teto: as I understand it, we use stub libraries when the libraries we would link against aren't available -- for example, because they exist outside the sandbox (like libcuda.so does, as part of the NVIDIA driver, in /run/opengl-driver/lib/). They allow the build to succeed where they would otherwise fail due to missing symbols.They shouldn't cause issues at runtime, because the executable should find and load the proper library from wherever it is it comes from (in this case, /run/opengl-driver/lib/). | 18:05:24 |
| teto | I dont seem to have any cuda library in /run/opengl-driver/. Should I add anything into hardware.opengl.extraPackages ? | 18:17:25 |
| connor (he/him) | You mean /run/opengl-driver/lib/ and not /run/opengl-driver/ right? | 18:41:09 |
| teto | I've searched both in depth so yes | 19:53:53 |
| connor (he/him) | What's the command you're using to try to run this piece of software? If it's a flake I can try to reproduce it on my machine | 20:37:10 |
 Mir | In reply to @ss:someonex.net Yeah, I had already figured out it but the bug issue is that I don't know what is the "right" way to include the definition of the half type. To make matters worse, I've tried to compile AliceVision on a docker container and using the official compilation scripts and yet the thing keeps failing. This means I can't even look at how the thing is supposed to compile. I'm at a bit of a loss as for how to proceed, but I suspect that I'll have to either ask the original authors for help or carefully read the cmake compilation scripts in order to look for potential sources of the error. Theoretically AliceVision has a nice CI pipeline but I can't see their build history so I don't even know how useful their CI scripts are. | 20:44:35 |
| teto | connor (he/him) (UTC-5): it's packaged in nixpkgs, nix run nixpkgs#local-ai (you need the override with config.cudaSupport true though). At one point I had GPU working but I use it on and off and now something changed in nixpkgs probably. | 21:11:33 |
| SomeoneSerge (matrix works sometimes) | In reply to @keiichi:matrix.org hardware.opengl.enable and the nvidia driver?.. | 21:29:16 |