| 28 Jan 2023 |
 hexa | did the same thing twice and it didnt | 20:46:50 |
| hexa | * did the same thing twice and it didn't work the second time either | 20:46:55 |
| hexa | it seems like patches in buildPythonPackage is too late? | 20:47:38 |
| hexa | which is surprising, since other dependencies are relaxed in postPatch there as well | 20:48:01 |
| hexa | oh, I think the path to the setup.py is different in there | 20:48:42 |
| hexa | but then it should complain about the patch not being applicable | 20:48:52 |
| hexa | tensorflow> applying patch /nix/store/7sgd7aq8bz4vyp9hz0vd02gizv69mar9-relax-python-deps.patch
tensorflow> patching file tensorflow/tools/pip_package/setup.py
| 20:49:08 |
| hexa | but in postPatch we patch ./setup.py | 20:49:41 |
| hexa | migrating to sed ._. | 20:50:08 |
| hexa | eh, used the time producitvely to write a module for tts-server 🙂 | 20:52:50 |
| hexa | builds on x86_64-linux | 21:02:12 |
| hexa | umap-learn fails to build as well 🙂 | 21:53:55 |
| hexa | because sklearn uses np.matrix with unsupported input 😐️ | 21:54:17 |
| hexa | Someone S: I guess we should merge tensorflow once I've confirmed it also builds on aarch64-linux, because even if it doesn't work for cuda or darwin we can fix that in another PR | 21:55:42 |
 SomeoneSerge (back on matrix) | Yes | 21:57:41 |
| hexa | sklearn updated, umap-learn fixed | 23:38:26 |
| 29 Jan 2023 |
| hexa | tf-cpu also builds on aarch64-darwin | 00:51:49 |
| SomeoneSerge (back on matrix) | I had switched hercules to use just one worker because I got annoyed by several simultaneous tensorflow builds throttling each. Now tf spends ages in the queue... | 02:47:30 |
| SomeoneSerge (back on matrix) | * I had switched hercules to use just one worker because I got annoyed by several simultaneous tensorflow builds throttling each other. Now tf spends ages in the queue... | 02:47:43 |
| 30 Jan 2023 |
| SomeoneSerge (back on matrix) | In reply to @hexa:lossy.network
tf-cpu also builds on aarch64-darwin python3.10-tensorflow-gpu-2.11.0 builds | 01:32:33 |
| hexa | awesome | 01:32:43 |
 Uri Baghin | tensorflow-probability bump if anyone's interested: https://github.com/NixOS/nixpkgs/pull/212421 | 10:19:13 |
| 2 Feb 2023 |
|  realjhol joined the room. | 21:32:33 |
| realjhol | hi - is it possible to install specific versions of tensorrt? I need to use v8.2.1.8, and I can't see any way to override the version in nixpkgs, or even use any version other than v8.4.0.6 | 21:34:45 |
 tpw_rules | https://github.com/NixOS/nixpkgs/blob/6462686d521c54d1adf405b026e9ad0f5a8d5c99/pkgs/development/libraries/science/math/tensorrt/extension.nix this is where it is defined. you'll probably want to change the file | 21:36:54 |
| realjhol | yes - I have that pulled up | 21:37:16 |
| realjhol | and there is the tensorRTVersions table | 21:37:26 |
| realjhol | -- which defines tensorrt versions and their compatibility with versions of cuda | 21:37:42 |
| realjhol | currently only 8.4.0 is supported in the table | 21:37:55 |
| realjhol | hypothetically if there was more than one supported version in the table, I can't see how to reference that derrivation | 21:38:25 |
