Upgrade FAISS to 1.7.x

[viclafargue]

Answers #3459

[dantegd]

Version in the conda recipe

cuml/conda/recipes/libcuml/meta.yaml

Line 43 in befdf98

- libfaiss=1.6.3

needs to be updated as well

Also we need a PR to the integration repo to update the version: https://github.com/rapidsai/integration, though before that pehaps it would be a good idea to use the ci/gpu/build.sh script

cuml/ci/gpu/build.sh

Line 62 in befdf98

# gpuci_conda_retry remove --force rapids-build-env rapids-notebook-env

to test CI in a commit before updating the integration repo

[JohnZed]

@viclafargue can you change this to target 0.19?

[JohnZed]

rerun tests

[viclafargue]

For some reason, with CUDA 11, conda installs libfaiss 1.6.3. There seems to be a conflict with the ucx/ucx-py packages.

[viclafargue]

libfaiss 1.7.0 cuda110h8045045_3_cuda
-------------------------------------
file name   : libfaiss-1.7.0-cuda110h8045045_3_cuda.tar.bz2
name        : libfaiss
version     : 1.7.0
build       : cuda110h8045045_3_cuda
build number: 3
size        : 88.0 MB
license     : MIT
subdir      : linux-64
url         : https://conda.anaconda.org/conda-forge/linux-64/libfaiss-1.7.0-cuda110h8045045_3_cuda.tar.bz2
md5         : b5a957da2eceab8e4e96fbfecdb51f6c
timestamp   : 2021-02-19 18:45:02 UTC
constraints : 
  - faiss-proc * cuda
dependencies: 
  - __glibc >=2.17
  - _openmp_mutex >=4.5
  - cudatoolkit 11.0|11.0.*
  - libblas >=3.8.0,<4.0a0
  - libgcc-ng >=7.5.0
  - liblapack >=3.8.0,<4.0a0
  - libstdcxx-ng >=7.5.0

ucx 1.8.1+g6b29558 cuda11.0_0
-----------------------------
file name   : ucx-1.8.1+g6b29558-cuda11.0_0.tar.bz2
name        : ucx
version     : 1.8.1+g6b29558
build       : cuda11.0_0
build number: 0
size        : 7.5 MB
license     : BSD-3-Clause
subdir      : linux-64
url         : https://conda.anaconda.org/rapidsai/linux-64/ucx-1.8.1+g6b29558-cuda11.0_0.tar.bz2
md5         : 95422ca7e866294c179d992dbc9e4320
timestamp   : 2020-08-26 22:23:57 UTC
constraints : 
  - ucx-proc * gpu
dependencies: 
  - __glibc >=2.17,<3.0.a0
  - _openmp_mutex >=4.5
  - cudatoolkit 11.0|11.0.*
  - libgcc-ng >=7.5.0
  - libstdcxx-ng >=7.5.0

It doesn't look like there should be any conflict. The installation works on my machine. I'll rerun the tests.

[viclafargue]

rerun tests

[dantegd]

@gpucibot merge

[dantegd]

rerun tests

[codecov-io]

Codecov Report

Merging #3509 (e3bd4d5) into branch-0.19 (a3bfb36) will decrease coverage by 35.38%.
The diff coverage is n/a.

@@               Coverage Diff                @@
##           branch-0.19    #3509       +/-   ##
================================================
- Coverage        80.69%   45.30%   -35.39%     
================================================
  Files              227      223        -4     
  Lines            17615    17045      -570     
================================================
- Hits             14214     7722     -6492     
- Misses            3401     9323     +5922     

Flag	Coverage Δ
dask	45.30% <ø> (+0.31%)	⬆️
non-dask	?

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
python/cuml/_thirdparty/__init__.py	0.00% <0.00%> (-100.00%)	⬇️
python/cuml/thirdparty_adapters/__init__.py	0.00% <0.00%> (-100.00%)	⬇️
python/cuml/_thirdparty/sklearn/exceptions.py	0.00% <0.00%> (-100.00%)	⬇️
python/cuml/experimental/explainer/__init__.py	0.00% <0.00%> (-100.00%)	⬇️
python/cuml/_thirdparty/sklearn/utils/__init__.py	0.00% <0.00%> (-100.00%)	⬇️
python/cuml/experimental/linear_model/__init__.py	0.00% <0.00%> (-100.00%)	⬇️
python/cuml/experimental/preprocessing/__init__.py	0.00% <0.00%> (-100.00%)	⬇️
...ython/cuml/experimental/hyperopt_utils/__init__.py	0.00% <0.00%> (-100.00%)	⬇️
...cuml/_thirdparty/sklearn/preprocessing/__init__.py	0.00% <0.00%> (-100.00%)	⬇️
python/cuml/benchmark/algorithms.py	0.00% <0.00%> (-94.39%)	⬇️
... and 129 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a3bfb36...e3bd4d5. Read the comment docs.