| 26 Jan 2024 |
 CRTified | In reply to @benoitdr:matrix.org
Is everyone OK with that ? I guess uploading large datasets to the nix store will increase the nix store hosting costs (on S3 ?). Another option could be to store the datasets in dedicated platforms (HugggingFace, Kaggle, ....) and store pointers in the nix store. Depending on how you point to the dataset, it will end up in the nix store after pulling | 11:15:51 |
| CRTified | Might require to break the sandboxing to allow network access | 11:16:10 |
 maupind | Does anyone have any experience overwriting python dependencies? I'm trying to get a project with botorch/ax working, but they have the package linear operator as a dependency which is failing due to its typeguard dependency being far behind. I feel like I am close with a post patch override but something is missing | 15:27:38 |
 raitobezarius | In reply to @benoitdr:matrix.org
Is everyone OK with that ? I guess uploading large datasets to the nix store will increase the nix store hosting costs (on S3 ?). Another option could be to store the datasets in dedicated platforms (HugggingFace, Kaggle, ....) and store pointers in the nix store. OK with what? | 15:32:39 |
| raitobezarius | We were not talking about cache.nixos.org, were we? | 15:32:58 |
 benoitdr | In reply to @raitobezarius:matrix.org
OK with what? OK with storing datasets in the nix store | 15:33:30 |
| raitobezarius | Well whether you upload it to the Nix store or to some random platform, what does that change? | 15:34:11 |
| raitobezarius | It needs to be stored somewhere? | 15:34:16 |
| benoitdr | In reply to @raitobezarius:matrix.org
Well whether you upload it to the Nix store or to some random platform, what does that change? the one who is paying for the storage ;-) There is no free lunch ... | 15:34:54 |
| raitobezarius | Again, we are not talking about cache.nixos.org, are we? | 15:35:28 |
| raitobezarius | Like even if you have pointers inside your remote storage, that does not change a lot that you are going to store a large buffer at some point somewhere, either in memory or on disk | 15:35:53 |
| benoitdr | In reply to @raitobezarius:matrix.org
Again, we are not talking about cache.nixos.org, are we? that's what I understood what maybe I got it wrong | 15:36:08 |
| raitobezarius | "The Nix store" is not cache.nixos.org, right? | 15:36:32 |
| benoitdr | In reply to @raitobezarius:matrix.org
Again, we are not talking about cache.nixos.org, are we? * that's what I understood but maybe I got it wrong | 15:36:44 |
| raitobezarius | Everyone who is using Nix has "the Nix store", in their local filesystem, it is in /nix/store | 15:36:46 |
| raitobezarius | But yes, storing datasets in cache.nixos.org is mostly out of the question, at the moment, for the cost reasons you mentioned | 15:37:12 |
| raitobezarius | Storing datasets in your own Nix store seems an interesting question to me though | 15:37:20 |
| raitobezarius | Compared to… store it locally on-disk, mount it from a remote location and buffer it, etc. | 15:37:44 |
| benoitdr | fully agree on that ... trexd , Can you clarify your request ? | 15:38:27 |
 trexd | In reply to @benoitdr:matrix.org
fully agree on that ... trexd , Can you clarify your request ? I just meant /nix/store locally. | 15:48:04 |
| benoitdr | OK sorry for confusion, personally I'll stick to a local minio server to store and share datasets locally. I tend to prefer a clear separation between code and data ;-) | 15:51:11 |
| benoitdr | I have built a docker image of an application that requires nvidia/cuda using dockerTools.buildImage. The image runs fine with nvidia-docker -run .... however, when I start a container from the same image via docker-compose, I have the following error :
Error response from daemon: failed to create task for container: failed to create shim task: OCI runtime create failed: runc create failed: unable to start container process: error during container init: error running hook #0: error running hook: exit status 1, stdout: , stderr: nvidia-container-cli: ldcache error: process /nix/store/cx01qk0qyylvkgisbwc7d3pk8sliccgh-glibc-2.38-27-bin/bin/ldconfig failed with error code: 1: unknown
I have added the few lines that are usually required to run nvidia/cuda images with docker-compose :
deploy:
resources:
reservations:
devices:
- driver: nvidia
device_ids: ['0']
capabilities: [gpu]
Note that other images (from docker hub) that require nvidia/cuda are working fine with these few lines.
Any idea what could go wrong ?
| 20:49:40 |
 SomeoneSerge (hash-versioned python modules when) | didn't read but do check out the nvidia-container-toolkit cdi PR maybe | 21:50:43 |
| 29 Jan 2024 |
|  bgrayburn joined the room. | 16:36:08 |
| 31 Jan 2024 |
|  Federico Damián Schonborn changed their profile picture. | 03:36:07 |
|  Alex Ou joined the room. | 04:57:44 |
| Federico Damián Schonborn changed their profile picture. | 06:21:45 |
| 3 Feb 2024 |
|  Tanja (Old; I'm now @tanja:catgirl.cloud) joined the room. | 02:50:47 |
| 5 Feb 2024 |
|  bbenno joined the room. | 13:07:46 |
| 9 Feb 2024 |
|  Daniel Fahey joined the room. | 10:32:08 |
