Open
Description
🚀 The feature
Right now the first time a user needs to run torchserve they need to
python install_dependencies.py
pip install torchserve or pip install .
This is confusing to users mostly used to the python ecosystem so how can we fix this? Especially as we move towards a python client like with the recent PR #2751 how can we make it easier to install torchserve within something like a notebook
Our install_dependencies.py is primarily used to install java, nodejs (for dev builds) and wget
We can't completely get rid of the script but we can recommend more users use conda if we think the below is more convenient
Motivation, pitch
The issue is torchserve is not a pure python library and at least within the pip ecosystem there is no way around this, you can't package non python dependencies easily
But you can in conda!
And I have a proof of concept, given this file binaries/conda/torchserve/meta.yaml where all we need to do is add our dependencies under run in the meta.yaml
package:
name: torchserve
version: "1.0"
source:
path: "/Users/marksaroufim/Dev/serve"
requirements:
build:
- python
- setuptools
run:
- python
- pillow
- psutil
- nodejs
- openjdk
- wget
build:
script: "{{ PYTHON }} -m pip install . -vv"
about:
home: https://github.com/pytorch/serve
summary: 'Model serving on PyTorch'
Build it
marksaroufim@marksaroufim-mbp conda % pwd
/Users/marksaroufim/Dev/serve/binaries/conda
marksaroufim@marksaroufim-mbp conda % conda build torchserve
Test it
marksaroufim@marksaroufim-mbp conda % conda create -n test-torchserve torchserve --use-local
Collecting package metadata (current_repodata.json): done
Solving environment: done
==> WARNING: A newer version of conda exists. <==
current version: 23.1.0
latest version: 23.9.0
Please update conda by running
$ conda update -n base -c defaults conda
Or to minimize the number of packages updated during conda update use
conda install conda=23.9.0
## Package Plan ##
environment location: /Users/marksaroufim/opt/anaconda3/envs/test-torchserve
added / updated specs:
- torchserve
The following packages will be downloaded:
package | build
---------------------------|-----------------
libdeflate-1.17 | hb664fd8_1 67 KB
libtiff-4.5.1 | hcec6c5f_0 508 KB
libuv-1.44.2 | h6c40b1e_0 388 KB
libwebp-1.3.2 | hf6ce154_0 84 KB
libwebp-base-1.3.2 | h6c40b1e_0 371 KB
nodejs-18.16.0 | he8a3799_1 11.8 MB
openjdk-11.0.13 | h8346a28_0 166.6 MB
pillow-10.0.1 | py39h7d39338_0 701 KB
torchserve-1.0 | py39_0 22.7 MB local
wget-1.21.4 | hf20ceda_1 364 KB
------------------------------------------------------------
Total: 203.5 MB
The following NEW packages will be INSTALLED:
ca-certificates pkgs/main/osx-64::ca-certificates-2023.08.22-hecd8cb5_0
freetype pkgs/main/osx-64::freetype-2.12.1-hd8bbffd_0
giflib pkgs/main/osx-64::giflib-5.2.1-h6c40b1e_3
icu pkgs/main/osx-64::icu-73.1-hcec6c5f_0
jpeg pkgs/main/osx-64::jpeg-9e-h6c40b1e_1
lcms2 pkgs/main/osx-64::lcms2-2.12-hf1fd2bf_0
lerc pkgs/main/osx-64::lerc-3.0-he9d5cce_0
libcxx pkgs/main/osx-64::libcxx-14.0.6-h9765a3e_0
libdeflate pkgs/main/osx-64::libdeflate-1.17-hb664fd8_1
libffi pkgs/main/osx-64::libffi-3.4.4-hecd8cb5_0
libidn2 pkgs/main/osx-64::libidn2-2.3.4-h6c40b1e_0
libpng pkgs/main/osx-64::libpng-1.6.39-h6c40b1e_0
libtiff pkgs/main/osx-64::libtiff-4.5.1-hcec6c5f_0
libunistring pkgs/main/osx-64::libunistring-0.9.10-h9ed2024_0
libuv pkgs/main/osx-64::libuv-1.44.2-h6c40b1e_0
libwebp pkgs/main/osx-64::libwebp-1.3.2-hf6ce154_0
libwebp-base pkgs/main/osx-64::libwebp-base-1.3.2-h6c40b1e_0
lz4-c pkgs/main/osx-64::lz4-c-1.9.4-hcec6c5f_0
ncurses pkgs/main/osx-64::ncurses-6.4-hcec6c5f_0
nodejs pkgs/main/osx-64::nodejs-18.16.0-he8a3799_1
openjdk pkgs/main/osx-64::openjdk-11.0.13-h8346a28_0
openjpeg pkgs/main/osx-64::openjpeg-2.4.0-h66ea3da_0
openssl pkgs/main/osx-64::openssl-3.0.11-hca72f7f_2
pillow pkgs/main/osx-64::pillow-10.0.1-py39h7d39338_0
pip pkgs/main/osx-64::pip-23.3-py39hecd8cb5_0
psutil pkgs/main/osx-64::psutil-5.9.0-py39hca72f7f_0
python pkgs/main/osx-64::python-3.9.18-h5ee71fb_0
readline pkgs/main/osx-64::readline-8.2-hca72f7f_0
setuptools pkgs/main/osx-64::setuptools-68.0.0-py39hecd8cb5_0
sqlite pkgs/main/osx-64::sqlite-3.41.2-h6c40b1e_0
tk pkgs/main/osx-64::tk-8.6.12-h5d9f67b_0
torchserve Users/marksaroufim/opt/anaconda3/conda-bld/osx-64::torchserve-1.0-py39_0
tzdata pkgs/main/noarch::tzdata-2023c-h04d1e81_0
wget pkgs/main/osx-64::wget-1.21.4-hf20ceda_1
wheel pkgs/main/osx-64::wheel-0.41.2-py39hecd8cb5_0
xz pkgs/main/osx-64::xz-5.4.2-h6c40b1e_0
zlib pkgs/main/osx-64::zlib-1.2.13-h4dc903c_0
zstd pkgs/main/osx-64::zstd-1.5.5-hc035e20_0
Proceed ([y]/n)?
You can see that nodejs, java and wget are all installed
Alternatives
No response
Additional context
No response