| 2 Dec 2024 |
 Gaétan Lepage | In reply to @glepage:matrix.org
Should be fixed by https://github.com/NixOS/nixpkgs/pull/361008. I went through both tensordict and torchrl. They should be more robust now. | 14:06:35 |
| Gaétan Lepage | Well, it's still not perfect hexa (UTC+1).
On the flaky Threadripper 3990X system, torchrl segfaults after the test suite has passed. It only occurs for python312Packages.torchrl but not for python311Packages.torchrl... | 14:40:15 |
| Gaétan Lepage | 
Download clipboard.png | 14:40:51 |
| Gaétan Lepage | Of course, it builds perfectly fine on two other AMD systems (Ryzen 5 5600X and Ryzen 9 3900) | 14:41:38 |
 hexa | microcode updates applied? | 14:42:05 |
| Gaétan Lepage | I forgot how to check that | 14:42:35 |
| Gaétan Lepage | I think you told me a while ago | 14:42:47 |
| hexa | intel be like | 14:43:46 |
| hexa |
[ 0.000000] microcode: updated early: 0xc6 -> 0xf8, date = 2024-02-01
| 14:43:51 |
| hexa | dmesg | grep microcode | 14:44:36 |
| Gaétan Lepage | [ 2.153548] microcode: microcode updated early to new patch_level=0x0830107b
| 14:45:07 |
| Gaétan Lepage | Nothing matches on my Ryzen system. | 14:46:03 |
| Gaétan Lepage | * TR system:
[ 2.153548] microcode: microcode updated early to new patch_level=0x0830107b
| 14:46:09 |
| Gaétan Lepage | * TR system:
[ 2.153548] microcode: microcode updated early to new patch_level=0x0830107b
...
[ 2.154392] microcode: Microcode Update Driver: v2.2.
| 14:46:30 |
| hexa | ❯ ./amd_ucode_info.py kernel/x86/microcode/AuthenticAMD.bin
Microcode patches in kernel/x86/microcode/AuthenticAMD.bin:
Family=0x10 Model=0x02 Stepping=0x03: Patch=0x01000083 Length=960 bytes
Family=0x10 Model=0x02 Stepping=0x02: Patch=0x01000083 Length=960 bytes
Family=0x10 Model=0x02 Stepping=0x0a: Patch=0x01000084 Length=960 bytes
Family=0x10 Model=0x06 Stepping=0x02: Patch=0x010000c7 Length=960 bytes
Family=0x10 Model=0x04 Stepping=0x03: Patch=0x010000c8 Length=960 bytes
Family=0x10 Model=0x06 Stepping=0x03: Patch=0x010000c8 Length=960 bytes
Family=0x10 Model=0x05 Stepping=0x03: Patch=0x010000c8 Length=960 bytes
Family=0x10 Model=0x08 Stepping=0x01: Patch=0x010000d9 Length=960 bytes
Family=0x10 Model=0x09 Stepping=0x01: Patch=0x010000d9 Length=960 bytes
Family=0x10 Model=0x08 Stepping=0x00: Patch=0x010000da Length=960 bytes
Family=0x10 Model=0x04 Stepping=0x02: Patch=0x010000db Length=960 bytes
Family=0x10 Model=0x05 Stepping=0x02: Patch=0x010000db Length=960 bytes
Family=0x10 Model=0x0a Stepping=0x00: Patch=0x010000dc Length=960 bytes
Family=0x11 Model=0x03 Stepping=0x01: Patch=0x02000032 Length=512 bytes
Family=0x12 Model=0x01 Stepping=0x00: Patch=0x03000027 Length=960 bytes
Family=0x14 Model=0x01 Stepping=0x00: Patch=0x05000029 Length=1568 bytes
Family=0x14 Model=0x02 Stepping=0x00: Patch=0x05000119 Length=1568 bytes
Microcode patches in kernel/x86/microcode/AuthenticAMD.bin+0x318c:
Family=0x15 Model=0x01 Stepping=0x02: Patch=0x0600063e Length=2592 bytes
Family=0x15 Model=0x02 Stepping=0x00: Patch=0x06000852 Length=2592 bytes
Family=0x15 Model=0x10 Stepping=0x01: Patch=0x06001119 Length=2592 bytes
Microcode patches in kernel/x86/microcode/AuthenticAMD.bin+0x5050:
Family=0x16 Model=0x00 Stepping=0x01: Patch=0x0700010f Length=3458 bytes
Microcode patches in kernel/x86/microcode/AuthenticAMD.bin+0x5e06:
Family=0x17 Model=0x01 Stepping=0x02: Patch=0x0800126f Length=3200 bytes
Family=0x17 Model=0x31 Stepping=0x00: Patch=0x0830107c Length=3200 bytes
Family=0x17 Model=0x08 Stepping=0x02: Patch=0x0800820d Length=3200 bytes
Family=0x17 Model=0xa0 Stepping=0x00: Patch=0x08a00008 Length=3200 bytes
Microcode patches in kernel/x86/microcode/AuthenticAMD.bin+0x9082:
Family=0x19 Model=0x01 Stepping=0x00: Patch=0x0a00107a Length=5568 bytes
Family=0x19 Model=0x11 Stepping=0x02: Patch=0x0a101248 Length=5568 bytes
Family=0x19 Model=0xa0 Stepping=0x02: Patch=0x0aa00215 Length=5568 bytes
Family=0x19 Model=0x01 Stepping=0x02: Patch=0x0a001238 Length=5568 bytes
Family=0x19 Model=0x11 Stepping=0x01: Patch=0x0a101148 Length=5568 bytes
Family=0x19 Model=0x01 Stepping=0x01: Patch=0x0a0011d5 Length=5568 bytes
Family=0x19 Model=0xa0 Stepping=0x01: Patch=0x0aa00116 Length=5568 bytes
| 14:52:33 |
| hexa | this is the latest ucode for amd on nixpkgs master | 14:52:39 |
| hexa | so 0x19 is family 25 | 14:53:59 |
| hexa | and for the model you probably have to binary or your model with 0xa0 if it is > 17 | 14:54:27 |
| Gaétan Lepage | Ok, so anyway the issues are not a problem of the python package then | 15:11:51 |
| 4 Dec 2024 |
 Kevin Mittman (UTC-8) | Anyone planning to attend PlanetNix https://www.socallinuxexpo.org/scale/22x/events/planet-nix ? Looks like CFP is still open | 02:21:01 |
 connor (he/him) | I probably will; I'm also planning to submit two talks | 04:38:33 |
| 6 Dec 2024 |
|  vannagamma joined the room. | 00:01:17 |
| connor (he/him) | Does anyone have a NixOS system they recommend using to test eval performance? | 05:49:07 |
| connor (he/him) | Ideally something which takes on the order of 30s or so to eval | 05:51:25 |
|  kaya 𖤐 changed their profile picture. | 21:17:32 |
| 7 Dec 2024 |
 SomeoneSerge (back on matrix) | In reply to @connorbaker:matrix.org
Does anyone have a NixOS system they recommend using to test eval performance? Working on cppnix? | 01:40:03 |
| connor (he/him) | Kind of? More like preliminary work for a talk I plan to give at Nix Planet | 06:52:28 |
| SomeoneSerge (back on matrix) | Looking forward to watch the record:) | 12:28:11 |
| 8 Dec 2024 |
| kaya 𖤐 | Hi, im attempting to upstream my nix derivation for exllamav2 from https://github.com/BatteredBunny/nix-ai-stuff/blob/main/pkgs/exllamav2.nix
For some reason its complaining about CUDA_HOME being missing even though im specifying it which im kind of confused about, i thought maybe i would replace torch with torchWithCuda but then i get some mysterious error which i dont get in the flake
Anyone had issues with anything similar? Current attempt for anyone curious https://gist.github.com/BatteredBunny/2212ac469f07244d954bf556f128cb07 | 16:42:14 |
| kaya 𖤐 | Pretty sure the biggest difference between the flake and my upstreaming attempt is that in the flake i have allowUnfree and allowUnfree as true, but those options should carry over (i assume) as i also have those enabled on my nixos and im building with --impure | 16:44:00 |
