| 27 Aug 2025 |
 nbp | https://bugzilla.mozilla.org/show_bug.cgi?id=1985439#c1 seeing that as part of my bugmail …
This sounds like a very bad news for Nix, given that I tried llama.cpp with various backends on my last computer, to find the ones which run and the one which perform well. | 16:03:08 |
 hexa | huh? | 16:04:50 |
| nbp | llama.cpp might have a way to select between various backend, but last time I tried this was a nightmare to get everything at once. (i.e. it did not compile) | 16:07:33 |
| nbp | and if Nvidia backend is enabled, we might have yet another build of Firefox. | 16:08:04 |
 aloisw | From the package definition it looks like you can build it without all of those and it will use BLAS, is that not good enough for Firefox? | 16:10:53 |
| aloisw | I don't think I'm the only one who doesn't want there Firefox package to depend on CUDA and ROCm for features they don't even use. | 16:11:37 |
| nbp | The problem is that we would have 3 variants of Firefox then, one optimized with Cuda, one with ROCm, and one without?
Sounds like we might want some re-linking mechanism to substitute one llama-cpp by another on a completed build of Firefox. | 16:14:32 |
 K900 | Can we just let it use vulkan compute or opencl or whatever generic shit it has | 16:19:38 |
| aloisw | It has all of them. | 16:19:59 |
| K900 | Cool can we just have it use that | 16:20:09 |
| K900 | Until someone complains | 16:20:12 |
| K900 | And then we tell them to fuck off and override llama-cpp if they want | 16:20:20 |
| aloisw | aloisw@exodus ~> nix path-info -S /nix/store/cwfaak5cpb6s49g427a4x9agxxgd1djc-llama-cpp-6210
/nix/store/cwfaak5cpb6s49g427a4x9agxxgd1djc-llama-cpp-6210 201550672
aloisw@exodus ~> nix path-info -S /nix/store/9pjd655imkq4qf0vv31ad80jhmszqh8s-llama-cpp-6210
/nix/store/9pjd655imkq4qf0vv31ad80jhmszqh8s-llama-cpp-6210 127200088
aloisw@exodus ~> nix path-info -S /nix/store/2z14x11p6l8wrbk7zjilid3259qgqjrm-llama-cpp-6210
/nix/store/2z14x11p6l8wrbk7zjilid3259qgqjrm-llama-cpp-6210 94045536
Wat, first is CPU, second is Vulkan, third is OpenCL. | 16:25:14 |
| 29 Aug 2025 |
|  @luna-null:matrix.org left the room. | 02:40:17 |
| nbp | https://blog.mozilla.org/en/firefox/firefox-ai/speeding-up-firefox-local-ai-runtime/ onnxruntime dependency explained | 10:26:38 |
| hexa | tried tab grouping recently and it was indeed super snappy | 13:18:22 |
| hexa | 
Download image.png | 13:33:02 |
| K900 | Hmm I wonder if the -bin packages need to be fixed for this too | 14:09:38 |
| K900 | I did get some "AI" popup in nightly the other day | 14:09:43 |
| 2 Sep 2025 |
| hexa | Also on 142.0 | 01:00:38 |
| 3 Sep 2025 |
|  ghpzin joined the room. | 07:55:19 |
| 5 Sep 2025 |
 vcunat | Firefox 145 will not have 32-bit Linux support. | 16:00:40 |
| vcunat | Though in nixpkgs it didn't build for a while already. | 16:01:37 |
 emily | I guess Chromium must still support it, as Steam uses Chromium? 🤔 | 16:04:50 |
| K900 | Steam uses 64-bit Chromium | 16:06:29 |
| emily | wait so Steam isn't actually 32 bits? | 16:06:56 |
| K900 | The main executable is | 16:08:16 |
| K900 | The Chrome bits are not | 16:08:19 |
| emily | I see | 16:09:04 |
| hexa | Redacted or Malformed Event | 19:25:26 |
