I am trying to install the pytorch-gpu in my Python-Transform repository, but it fails each time giving an error. but on the other hand, the pytorch package is getting successfully installed.
i am attaching the meta.yaml for current config and the screenshots of the error, i even checked on the artifacts page, it shows a lot of versions for pytorch-gpu are unavailable, i wonder if this is the issue for it not getting installed successfully, can someone please help figure out what the problem must be?
this is my current meta.yaml file
If you need to modify the runtime requirements for your package,
# update the ‘requirements.run’ section in this file.
package:
name: “{{ PACKAGE_NAME }}”
version: “{{ PACKAGE_VERSION }}”
source:
path: ../src
requirements:
pip:
- uuid
- paddleocr
- paddlepaddle
build:
- python
- setuptools
run:
- python
- transforms {{ PYTHON_TRANSFORMS_VERSION }}
- transforms-expectations
- transforms-preview
- transforms-verbs
- foundry-transforms-lib-python
- palantir_models
- transforms-external-systems
- libglib
- libopencv
- libgomp
- google-cloud-vision
- transforms-media
- libcblas
- setuptools
- pandas
- transformers
- torchvision
- python-dateutil
- numpy
- pymupdf
- pillow
- groq
- typing-extensions
- cryptography
- ultralytics
test:
requires:
build:
script: python setup.py install --single-version-externally-managed --record=record.txt
I ran into this issue as well. It’s crazy that this issue was reported 6 months ago and no one responded. I changed the config to this
# If you need to modify the runtime requirements for your package,
# update the 'requirements.run' section in this file.
package:
name: "{{ PACKAGE_NAME }}"
version: "{{ PACKAGE_VERSION }}"
source:
path: ../src
requirements:
# Tools required to build the package. These packages are run on the build system and include
# things such as revision control systems (Git, SVN) make tools (GNU make, Autotool, CMake) and
# compilers (real cross, pseudo-cross, or native when not cross-compiling), and any source pre-processors.
# https://docs.conda.io/projects/conda-build/en/latest/resources/define-metadata.html#build
build:
- python
- setuptools
# Packages required to run the package. These are the dependencies that are installed automatically
# whenever the package is installed.
# https://docs.conda.io/projects/conda-build/en/latest/resources/define-metadata.html#run
run:
- python
- transforms
- transforms-expectations
- transforms-preview
- transforms-verbs
- foundry-transforms-lib-python
- aip-workflows
test:
requires:
- transforms 3.193.0
build:
script: python setup.py install --single-version-externally-managed --record=record.txt
noarch: python
Committed the change, then refreshed the page (don’t just rebuild workspace) and it worked.
Hello,
I was having similar problems but following the documentation solved it for me:
Developer guides • Train models with GPUs • Palantir
You need to add a code snippet like below into the most nested build.grade of your Code Repository. This should solve the __cuda dependency error that you might be encountering when installing pytorch-gpu
// CUDA setup
tasks.runVersions {
// 12.8 is just an example, use the latest supported
// CUDA version for your model training framework version
// and GPU/driver.
// See https://www.tensorflow.org/install/source#gpu for tensorflow
// and https://pytorch.org/get-started/locally/ under linux for pytorch.
environment(“CONDA_OVERRIDE_CUDA”, “12.8”)
}
tasks.condaPackRun {
environment(“CONDA_OVERRIDE_CUDA”, “12.8”)
}
