| 6 Mar 2023 |
 SomeoneSerge (matrix works sometimes) | .github/workflows/_linux-build.yml
161: docker exec -t "${container_name}" sh -c '.jenkins/pytorch/build.sh'
| 22:45:01 |
| SomeoneSerge (matrix works sometimes) | Yes, I think this is the right one: https://github.com/pytorch/pytorch/blob/39e8311a29b5713c8858cab73a8f713a7f3d531c/.github/workflows/_binary-build-linux.yml#L205
...but they still take the flags from elsewhere and just propagate them | 22:51:02 |
| SomeoneSerge (matrix works sometimes) | aaaaaand 0 workflows run https://github.com/pytorch/pytorch/actions/workflows/_binary-build-linux.yml | 22:51:55 |
 hexa | yeah, why would they run that 😄 | 22:52:07 |
| SomeoneSerge (matrix works sometimes) | https://github.com/pytorch/pytorch/actions/runs/4337823562 | 22:54:01 |
| SomeoneSerge (matrix works sometimes) | Here https://github.com/pytorch/pytorch/actions/runs/4337823562/jobs/7574087583#step:14:305 | 22:55:00 |
| SomeoneSerge (matrix works sometimes) | -DBUILD_LIBTORCH_CPU_WITH_DEBUG=0
Ok, how do we check we don't have any debug symbols in our libs?
| 22:55:59 |
| hexa | objdump --syms | 22:59:10 |
| SomeoneSerge (matrix works sometimes) | -DUSE_NCCL=1 | 22:59:17 |
| SomeoneSerge (matrix works sometimes) | hmmm, I didn't even know it can be built without cuda | 23:00:18 |
| SomeoneSerge (matrix works sometimes) | In reply to @hexa:lossy.network
objdump --syms Seems fine | 23:04:58 |
| hexa | agreed | 23:05:05 |
| SomeoneSerge (matrix works sometimes) | -DUSE_FBGEMM? | 23:07:02 |
| hexa |
FBGEMM (Facebook GEneral Matrix Multiplication) is a low-precision, high-performance matrix-matrix multiplications and convolution library
| 23:09:25 |
| SomeoneSerge (matrix works sometimes) | ❯ nix log nixpkgs#python3Packages.torch
...
-- USE_EIGEN_FOR_BLAS : ON
-- USE_FBGEMM : ON
-- USE_FAKELOWP : OFF
-- USE_KINETO : ON
...
| 23:09:47 |
| 7 Mar 2023 |
| SomeoneSerge (matrix works sometimes) | (how does one get an nvim diffsplit online?) | 00:01:28 |
| SomeoneSerge (matrix works sometimes) | Meanwhile, two notable GLIBCXX_3.4.30 failures remain: python3Packages.jax and python3Packages.torchvision | 00:39:55 |
| SomeoneSerge (matrix works sometimes) | Same error I saw with gpflow, something about scipy/optimize:
from ._highs._highs_wrapper import _highs_wrapper
E ImportError: /nix/store/205vsmbfhq1q2vhgskpqyymqvba4mscp-gcc-11.3.0-lib/lib/libstdc++.so.6: version `GLIBCXX_3.4.30' not found (required by /nix/store/yi7jc5p2mlwb3j37j7gwj15bk45j6xqs-python3.10-scipy-1.9.3/lib/python3.10/site-packages/scipy/optimize/_highs/_highs_wrapper.cpython-310-x86_64-linux-gnu.so)
| 00:40:34 |
| SomeoneSerge (matrix works sometimes) | * Meanwhile, two notable GLIBCXX_3.4.30 failures remain: python3Packages.jax (not jaxlib) and python3Packages.torchvision | 01:21:16 |
| SomeoneSerge (matrix works sometimes) | MKL -> down to 5s | 01:29:23 |
| SomeoneSerge (matrix works sometimes) | Mystery solved, I guess | 01:29:38 |
| SomeoneSerge (matrix works sometimes) | connor (he/him): you should've received an automated invite to the NixOS GH organization back when your first PR was merged. I don't see NixOS in your org list. Could you have missed it? | 12:07:02 |
| SomeoneSerge (matrix works sometimes) | * connor (he/him): you should've received an automated invite to the NixOS GH organization back when your first PR was merged. I don't see NixOS in your org list. Could you have missed it?
EDIT: Oh! That's how it works: https://github.com/NixOS/nixpkgs/issues/50105#issuecomment-1458140917
Add yourself to maintainers/maintainer-list.nix
| 13:16:35 |
| hexa | from where do I get nvcc? | 14:30:56 |
| hexa | nix-locate is being entirely unhelpful | 14:31:10 |
| hexa | I suspect somewhere below cudatoolkit? | 14:31:24 |
 Kevin Mittman (UTC-7) | cuda_nvcc ? | 15:08:10 |
| hexa | oh, thanks | 15:18:25 |
| hexa | a bit redundant, below cudaPackages | 15:18:36 |
| SomeoneSerge (matrix works sometimes) | In reply to @hexa:lossy.network
a bit redundant, below cudaPackages Yea, but the names come from the json manifests published by nvidia. And if we strip cuda_ from cuda_nvcc, should we keep lib in libcublas? And then you learn that libcusparse and cuSPARSE are different things | 16:30:18 |
