What exactly do you mean by 'manually remove from installed packages in the UI'? Where on the UI?

I dont know where cudatoolkit=12.2 is taken from. Its not on requirements.txt

on the virtual env, these are the installed packaged-

I tried adding
Task.add_requirements("cudatoolkit==12.2")#replacing pip install cudatoolkit==12.2

but then got
...
agent.package_manager.torch_nightly = false
agent.package_manager.poetry_files_from_repo_working_dir = false
agent.venvs_dir = /home/rakefet/.clearml/venvs-builds
agent.venvs_cache.max_entries = 10
agent.venvs_cache.free_space_threshold_gb = 2.0
agent.venvs_cache.path = ~/.clearml/venvs-cache
agent.vcs_cache.enabled = true
agent.vcs_cache.path = /home/rakefet/.clearml/vcs-cache
...

looking now i think it's probably since on my code I first read parameters from config file-
config = HpsYaml(paras.config)
and after that, I set these parameters to dictionary, which I connect to the task-
hparams_dict = {'batch_size': config.hparas.batch_size,
'valid_step': config.hparas.valid_step,
'max_step': config.hparas.max_step}
parameters = task.connect(hparams_dict, name='hyper_params')
So maybe the parameters from the config file override th...

Thanks

Also, I am indeed using conda as package manager.
package_manager: {
# supported options: pip, conda, poetry
type: conda,

As you mentioned, requirement for 'cudatoolkit=12.2' is internal to clearml-agent, so I have no access of how to solve it.

Im running of Dell XPS 15 7590 with OS Ubuntu 22.04.2 (not a mac)
proocessor - x86_64.
Did update but still getting same error

locally the virtual env is created with conda, but inside it there are also packages installed with pip. Is that what you mean?

This is holding me from proceeding for quite a long.. perhapse we can meet virtually and solve it?

on agent config I use conda , as I previously shared

But what about this error?
ERROR: Invalid requirement: 'cudatoolkit=12.2'
Hint: = is not a valid operator. Did you mean == ?
RequirementsManager handler
...
exception: Failed installing GIT/HTTPs package 'cudatoolkit=12.2'
Failed installing GIT/HTTPs package 'cudatoolkit=12.2'
clearml_agent: ERROR: Could not install task requirements!

What is the origin of cudatoolkit=12.2 ? How should I resolve it?

looking at 'installed packages' section after Taske reset I only see that ( NO cuda toolkit)-

Python 3.8.17 (default, Jul 5 2023, 21:04:15) [GCC 11.2.0]

AMFM_decompy == 1.0.11
Cython == 3.0.2
Pillow == 10.0.0
PyYAML == 6.0.1
bitarray == 2.8.1
clearml == 1.12.2
einops == 0.6.1
hydra_core == 1.0.7
joblib == 1.3.2
librosa == 0.10.1
matplotlib == 3.7.2
numpy == 1.24.4
omegaconf == 2.0.6
packaging == 23.1
psutil == 5.9.5
regex == 2023.8.8
requests == 2.31.0
sacrebleu == 2.3.1
scikit_learn == ...

the original Task is created by simply executing code, not through agent

can you suggest a solution or a workaround?

Hi! after a deeper check I realized that I had also problem on my local pc to communicate with Nvidia driver. I now re-installed driver and dependencies, validated with nvidia-smi command, and local run looks ok.
I re-run with clearml-agent, now getting thie error-
Successfully installed AMFM_decompy-1.0.11 MarkupSafe-2.1.3 Pillow-10.0.0 PyYAML-6.0.1 antlr4-python3-runtime-4.8 appdirs-1.4.4 attrs-23.1.0 audioread-3.0.0 bitarray-2.7.6 cffi-1.15.1 clearml-1.12.2 cmake-3.27.2 colorama-0.4.6 con...

Ok..so I should generally avoid connecting complex objects? I guess I would create a 'mini dictionary' with a subset of params, and connectvthis instead.

Ok. Thanks. will try

Hi, I rerun now after minor updates. Get similar error in the same part of code-
Traceback (most recent call last):
File "main.py", line 84, in <module>
task.connect(config.model,name='model params')
File "/home/rakefet/.clearml/venvs-builds/3.8/lib/python3.8/site-packages/clearml/task.py", line 1455, in connect
return method(mutable, name=name)
File "/home/rakefet/.clearml/venvs-builds/3.8/lib/python3.8/site-packages/clearml/task.py", line 3573, in _connect_dictionary
dicti...

What about the second issue I had regarding unrecognized package?

I managed to get to the stage where 'Environment setup completed successfully' for the enqueued task!
However, now I encountered another issue - an error that did not occur on the manual run.
Related to
task.connect(config.model,name='model params')
I wonder if this is related to environment

Hi, Thanks! I thought it would be good to use conda since I used it on my local machine , but I understand now pip can manage everything.

Just to be clear, regarding the task.connect I have no solution. I can't keep these line commented out. Tnx in advance

You are right! I added this and indeed issue was solved. Thanks!

Before doing anything I got -
Environment setup completed successfully
Starting Task Execution:
Traceback (most recent call last):
File "inference.py", line 10, in <module>
import soundfile as sf
File "/home/ubuntu/.clearml/venvs-builds/3.8/lib/python3.8/site-packages/soundfile.py", line 142, in <module>
raise OSError('sndfile library not found')
OSError: sndfile library not found

It may be related to the fact i re-installed cuda drivers. and did not re-create the virtual envs. However, on my pc it runs on gpu with no errors

Hi. To be on the safe side, I recreated the virtual env, ran locally and after through locally installed agent.
I get the same error - see log file.

After removing the task.connect lines, it encountered another error related to 'einops' that is not recognized. It does exist on my environment file but was not installed by the agent (according to what I see on 'Summary - installed python packages'. should I add this manually?