Skip to content

Commit 2964a37

Browse files
Michael Norrismeta-codesync[bot]
Michael Norris
authored and
meta-codesync[bot]
committed
Fix nightly by updating mkl version (#4604)
Summary: Pull Request resolved: #4604 Nightly for GPU builds (cuvs and non-cuvs) is failing again after numpy2 upgrade. - Fixed it by upgrading CUDA, using the right deps (pytorch-gpu instead of the deprecated pytorch-cuda), pinning pytorch >=2.7 for GPU builds, upgrading MKL - SVE requires pytorch<2.5 for the testing flow, otherwise it times out for 1 test. Reviewed By: subhadeepkaran Differential Revision: D84193438 fbshipit-source-id: 71868d5f9deb973ee431e10ee765a450d4f1f876
1 parent 752832c commit 2964a37

File tree

6 files changed

+37
-46
lines changed

6 files changed

+37
-46
lines changed

‎.github/actions/build_cmake/action.yml‎

Lines changed: 5 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -50,27 +50,28 @@ runs:
5050
if [ "${{ runner.arch }}" = "X64" ]; then
5151
# TODO: merge this with ARM64
5252
conda install -y -q -c conda-forge gxx_linux-64=14.2 sysroot_linux-64=2.17
53-
conda install -y -q mkl=2022.2.1 mkl-devel=2022.2.1
53+
conda install -y -q mkl=2024.2.2 mkl-devel=2024.2.2
5454
fi
5555
5656
# no CUDA needed for ROCm so skip this
5757
if [ "${{ inputs.rocm }}" = "ON" ]; then
5858
:
5959
# regular CUDA for GPU builds
6060
elif [ "${{ inputs.gpu }}" = "ON" ] && [ "${{ inputs.cuvs }}" = "OFF" ]; then
61-
conda install -y -q cuda-toolkit=12.4 -c "nvidia/label/cuda-12.4.0"
61+
conda install -y -q cuda-toolkit=12.6 gxx_linux-64=12.4 -c "nvidia/label/cuda-12.6"
6262
# and CUDA from cuVS channel for cuVS builds
6363
elif [ "${{ inputs.cuvs }}" = "ON" ]; then
64-
conda install -y -q libcuvs=25.08 'cuda-version>=12.0,<=12.5' cuda-toolkit=12.4.1 gxx_linux-64=12.4 -c rapidsai -c rapidsai-nightly -c conda-forge
64+
conda install -y -q libcuvs=25.08 'cuda-version=12.6' cuda-toolkit=12.6 gxx_linux-64=12.4 -c rapidsai -c rapidsai-nightly -c conda-forge
6565
fi
6666
6767
# install test packages
6868
if [ "${{ inputs.rocm }}" = "ON" ]; then
6969
: # skip torch install via conda, we need to install via pip to get
7070
# ROCm-enabled version until it's supported in conda by PyTorch
7171
elif [ "${{ inputs.gpu }}" = "ON" ]; then
72-
conda install -y -q "pytorch<2.5" pytorch-cuda=12.4 -c pytorch -c "nvidia/label/cuda-12.4.0"
72+
conda install -y -q "pytorch>=2.7" "pytorch-gpu>=2.7" -c pytorch -c "nvidia/label/12.6"
7373
else
74+
# TestLowLevelIVF.IVFRQ hangs on pytorch>=2.7, so left it as <2.5 for now.
7475
conda install -y -q "pytorch<2.5" -c pytorch
7576
fi
7677
- name: ROCm - Install dependencies

‎.github/workflows/build-release.yml‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -23,7 +23,7 @@ jobs:
2323
with:
2424
label: main
2525
linux-x86_64-GPU-packages-CUDA-12-4-0:
26-
name: Linux x86_64 GPU packages (CUDA 12.4.0)
26+
name: Linux x86_64 GPU packages (CUDA 12.6)
2727
runs-on: 4-core-ubuntu-gpu-t4
2828
env:
2929
CUDA_ARCHS: "70-real;72-real;75-real;80;86-real"
@@ -39,9 +39,9 @@ jobs:
3939
ANACONDA_API_TOKEN: ${{ secrets.ANACONDA_API_TOKEN }}
4040
with:
4141
label: main
42-
cuda: "12.4.0"
42+
cuda: "12.6"
4343
linux-x86_64-GPU-CUVS-packages-CUDA12-4-0:
44-
name: Linux x86_64 GPU w/ cuVS packages (CUDA 12.4.0)
44+
name: Linux x86_64 GPU w/ cuVS packages (CUDA 12.6)
4545
runs-on: 4-core-ubuntu-gpu-t4
4646
env:
4747
CUDA_ARCHS: "70-real;72-real;75-real;80;86-real"
@@ -58,7 +58,7 @@ jobs:
5858
with:
5959
label: main
6060
cuvs: "ON"
61-
cuda: "12.4.0"
61+
cuda: "12.6"
6262
windows-x86_64-packages:
6363
name: Windows x86_64 packages
6464
runs-on: windows-2022

‎.github/workflows/nightly.yml‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -21,7 +21,7 @@ jobs:
2121
with:
2222
label: nightly
2323
linux-x86_64-GPU-CUDA-12-4-nightly:
24-
name: Linux x86_64 GPU nightlies (CUDA 12.4.0)
24+
name: Linux x86_64 GPU nightlies (CUDA 12.6)
2525
runs-on: 4-core-ubuntu-gpu-t4
2626
env:
2727
CUDA_ARCHS: "70-real;72-real;75-real;80;86-real"
@@ -36,9 +36,9 @@ jobs:
3636
ANACONDA_API_TOKEN: ${{ secrets.ANACONDA_API_TOKEN }}
3737
with:
3838
label: nightly
39-
cuda: "12.4.0"
39+
cuda: "12.6"
4040
linux-x86_64-GPU-CUVS-CUDA12-4-0-nightly:
41-
name: Linux x86_64 GPU w/ cuVS nightlies (CUDA 12.4.0)
41+
name: Linux x86_64 GPU w/ cuVS nightlies (CUDA 12.6)
4242
runs-on: 4-core-ubuntu-gpu-t4
4343
env:
4444
CUDA_ARCHS: "70-real;72-real;75-real;80;86-real"
@@ -54,7 +54,7 @@ jobs:
5454
with:
5555
label: nightly
5656
cuvs: "ON"
57-
cuda: "12.4.0"
57+
cuda: "12.6"
5858
windows-x86_64-nightly:
5959
name: Windows x86_64 nightlies
6060
runs-on: windows-2022

‎INSTALL.md‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -36,7 +36,7 @@ $ conda install -c pytorch/label/nightly faiss-cpu
3636
# GPU(+CPU) version
3737
$ conda install -c pytorch/label/nightly -c nvidia faiss-gpu=1.12.0
3838

39-
# GPU(+CPU) version with NVIDIA cuVS (package built with CUDA 12.4)
39+
# GPU(+CPU) version with NVIDIA cuVS (package built with CUDA 12.6)
4040
conda install -c pytorch -c rapidsai -c rapidsai-nightly -c conda-forge -c nvidia pytorch/label/nightly::faiss-gpu-cuvs 'cuda-version>=12.0,<=12.5'
4141

4242
# GPU(+CPU) version with NVIDIA cuVS (package built with CUDA 11.8)

‎conda/faiss-gpu-cuvs/meta.yaml‎

Lines changed: 11 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -6,9 +6,9 @@
66
{% set version = environ.get('GIT_DESCRIBE_TAG').lstrip('v') %}
77
{% set suffix = "_nightly" if environ.get('PACKAGE_TYPE') == 'nightly' else "" %}
88
{% set number = GIT_DESCRIBE_NUMBER %}
9-
{% set cuda_constraints=">=12.1,<12.5" %}
10-
{% set libcublas_constraints=">=12.1,<13" %}
11-
{% set cudart_constraints=">=12.4,<12.5" %}
9+
{% set cuda_constraints=">=12.6,<12.7" %}
10+
{% set libcublas_constraints=">=12.6,<12.7" %}
11+
{% set cudart_constraints=">=12.6,<12.7" %}
1212

1313
package:
1414
name: faiss-pkg
@@ -47,8 +47,8 @@ outputs:
4747
- cmake >=3.30.4
4848
- make =4.2 # [not win]
4949
- _openmp_mutex =4.5=2_kmp_llvm # [x86_64]
50-
- mkl =2023 # [x86_64]
51-
- mkl-devel =2023 # [x86_64]
50+
- mkl >=2024.2.2,<2025.0a0 # [x86_64]
51+
- mkl-devel >=2024.2.2,<2025.0a0 # [x86_64]
5252
- cuda-toolkit {{ cudatoolkit }}
5353
- cuda-cudart {{ cudart_constraints }}
5454
- cuda-cudart-dev {{ cudart_constraints }}
@@ -58,13 +58,13 @@ outputs:
5858
- cuda-cudart-static_linux-64 {{ cudart_constraints }} # [linux64]
5959
host:
6060
- _openmp_mutex =4.5=2_kmp_llvm # [x86_64]
61-
- mkl =2023 # [x86_64]
61+
- mkl >=2024.2.2,<2025.0a0 # [x86_64]
6262
- openblas =0.3.30 # [not x86_64]
6363
- libcuvs =25.08
6464
- cuda-version {{ cuda_constraints }}
6565
run:
6666
- _openmp_mutex =4.5=2_kmp_llvm # [x86_64]
67-
- mkl =2023 # [x86_64]
67+
- mkl >=2024.2.2,<2025.0a0 # [x86_64]
6868
- openblas =0.3.30 # [not x86_64]
6969
- cuda-cudart {{ cuda_constraints }}
7070
- libcublas {{ libcublas_constraints }}
@@ -95,16 +95,16 @@ outputs:
9595
- cmake >=3.26.4
9696
- make =4.2 # [not win]
9797
- _openmp_mutex =4.5=2_kmp_llvm # [x86_64]
98-
- mkl =2023.0 # [x86_64]
98+
- mkl >=2024.2.2,<2025.0a0 # [x86_64]
9999
- cuda-toolkit {{ cudatoolkit }}
100100
host:
101-
- mkl =2023.0 # [x86_64]
101+
- mkl >=2024.2.2,<2025.0a0 # [x86_64]
102102
- _openmp_mutex =4.5=2_kmp_llvm # [x86_64]
103103
- python {{ python }}
104104
- numpy >=2.0,<3.0
105105
- {{ pin_subpackage('libfaiss', exact=True) }}
106106
run:
107-
- mkl =2023.0 # [x86_64]
107+
- mkl >=2024.2.2,<2025.0a0 # [x86_64]
108108
- _openmp_mutex =4.5=2_kmp_llvm # [x86_64]
109109
- python {{ python }}
110110
- numpy >=2.0,<3.0
@@ -114,8 +114,7 @@ outputs:
114114
requires:
115115
- numpy >=2.0,<3.0
116116
- scipy
117-
- pytorch >=2.7
118-
- pytorch-cuda {{ cuda_constraints }}
117+
- pytorch-gpu >=2.7
119118
commands:
120119
- python -X faulthandler -m unittest discover -v -s tests/ -p "test_*"
121120
- python -X faulthandler -m unittest discover -v -s tests/ -p "torch_*"

‎conda/faiss-gpu/meta.yaml‎

Lines changed: 12 additions & 21 deletions
Original file line numberDiff line numberDiff line change
@@ -6,16 +6,9 @@
66
{% set version = environ.get('GIT_DESCRIBE_TAG').lstrip('v') %}
77
{% set suffix = "_nightly" if environ.get('PACKAGE_TYPE') == 'nightly' else "" %}
88
{% set number = GIT_DESCRIBE_NUMBER %}
9-
{% if cudatoolkit == '11.4.4' %}
10-
{% set cuda_constraints=">=11.4,<12" %}
11-
{% set libcublas_constraints=">=11.6,<12" %}
12-
{% elif cudatoolkit == '12.1.1' %}
13-
{% set cuda_constraints=">=12.1,<13" %}
14-
{% set libcublas_constraints=">=12.1,<13" %}
15-
{% elif cudatoolkit == '12.4.0' %}
16-
{% set cuda_constraints=">=12.4,<13" %}
17-
{% set libcublas_constraints=">=12.4,<13" %}
18-
{% endif %}
9+
{% set cuda_constraints=">=12.6,<12.7" %}
10+
{% set libcublas_constraints=">=12.6,<12.7" %}
11+
{% set cudart_constraints=">=12.6,<12.7" %}
1912

2013
package:
2114
name: faiss-pkg
@@ -49,20 +42,19 @@ outputs:
4942
- FAISS_FLATTEN_CONDA_INCLUDES
5043
requirements:
5144
build:
52-
- {{ compiler('cxx') }}
45+
- {{ compiler('cxx') }} =12.4
5346
- sysroot_linux-64 =2.17 # [linux64]
5447
- llvm-openmp # [osx]
5548
- cmake >=3.24.0
5649
- make =4.2 # [not win and not (osx and arm64)]
5750
- make =4.4 # [osx and arm64]
58-
- mkl-devel =2023.0 # [x86_64]
51+
- mkl-devel >=2024.2.2,<2025.0a0 # [x86_64]
5952
- cuda-toolkit {{ cudatoolkit }}
60-
- gcc_linux-64 =11.2 # [cudatoolkit == '11.4.4']
6153
host:
62-
- mkl =2023.0 # [x86_64]
54+
- mkl >=2024.2.2,<2025.0a0 # [x86_64]
6355
- openblas =0.3.30 # [not x86_64]
6456
run:
65-
- mkl =2023.0 # [x86_64]
57+
- mkl >=2024.2.2,<2025.0a0 # [x86_64]
6658
- openblas =0.3.30 # [not x86_64]
6759
- cuda-cudart {{ cuda_constraints }}
6860
- libcublas {{ libcublas_constraints }}
@@ -84,23 +76,23 @@ outputs:
8476
string: "py{{ PY_VER }}_h{{ PKG_HASH }}_{{ number }}_cuda{{ cudatoolkit }}{{ suffix }}"
8577
requirements:
8678
build:
87-
- {{ compiler('cxx') }}
79+
- {{ compiler('cxx') }} =12.4
8880
- sysroot_linux-64 =2.17 # [linux64]
8981
- swig =4.0
9082
- cmake >=3.24.0
9183
- make =4.2 # [not win and not (osx and arm64)]
9284
- make =4.4 # [osx and arm64]
9385
- _openmp_mutex =4.5=2_kmp_llvm # [x86_64 and not win]
9486
- cuda-toolkit {{ cudatoolkit }}
95-
- mkl-devel =2023.0 # [x86_64]
87+
- mkl-devel >=2024.2.2,<2025.0a0 # [x86_64]
9688
host:
97-
- mkl =2023.0 # [x86_64]
89+
- mkl >=2024.2.2,<2025.0a0 # [x86_64]
9890
- python {{ python }}
9991
- numpy >=2.0,<3.0
10092
- _openmp_mutex =4.5=2_kmp_llvm # [x86_64 and not win]
10193
- {{ pin_subpackage('libfaiss', exact=True) }}
10294
run:
103-
- mkl =2023.0 # [x86_64]
95+
- mkl >=2024.2.2,<2025.0a0 # [x86_64]
10496
- python {{ python }}
10597
- numpy >=2.0,<3.0
10698
- packaging
@@ -109,8 +101,7 @@ outputs:
109101
requires:
110102
- numpy >=2.0,<3.0
111103
- scipy
112-
- pytorch >=2.7
113-
- pytorch-cuda {{ cuda_constraints }}
104+
- pytorch-gpu >=2.7
114105
commands:
115106
- python -X faulthandler -m unittest discover -v -s tests/ -p "test_*"
116107
- python -X faulthandler -m unittest discover -v -s tests/ -p "torch_*"

0 commit comments

Comments
 (0)