feat(logger.py): support swanlab

[tpoisonooo]

What does this PR do ?

Same as #691 , wandb is unreachable for our district, add swanlab by instead. Just like github --> gitee

Issues

List issues that this PR closes (syntax):

No

Usage

• You can potentially add a usage example below

Here is my example config:

logger:
  log_dir: "logs"  # Base directory for all logs
  num_val_samples_to_print: 0 # Number of validation samples to pretty print on terminal
  wandb_enabled: false
  swanlab_enabled: true
  tensorboard_enabled: true
  monitor_gpus: false  # If true, will monitor GPU usage and log to wandb and/or tensorboard
  wandb:
    project: "grpo-dev"
    name: "sj_megatron_1B"
  swanlab:
    project: "grpo-dev"
    name: "sj_megatron_1B"

Here is swanlab UI and public project

# Add a code snippet demonstrating how to use this

Before your PR is "Ready for review"

Pre checks:

• Make sure you read and followed Contributor guidelines
• Did you write any new necessary tests?
• Did you run the unit tests and functional tests locally? Visit our Testing Guide for how to run tests
• Did you add or update any necessary documentation? Visit our Document Development Guide for how to write, build and test the docs.

Additional Information

• ...

[tpoisonooo]

@terrykong
I already read #534 and update the code, but how to run the pytest.

When I uv run pytest tests/unit/utils/test_logger.py, it calls ray env and download some gated models. Looks like not a unittest.

run ray cluster

download some models

[terrykong]

@tpoisonooo can you try these instructions:

https://docs.nvidia.com/nemo/rl/latest/testing.html#unit-tests

So to just test the logger, i'd run:

uv run --group test bash tests/run_unit.sh -k test_logger

You may also run bare pytest, but we've had some issues where pytest was sensitive to the current working directory, so we have this script that can be run more robustly for convenience.

[SahilJain314]

I think you do actually need HF access to those models @terrykong, the unit tests use them (fwiw, I've been able to run bare pytest in the past so long as I have the HF key). Maybe we should switch fully to qwen models in the tests to make it easier for community members to run locally?

@tpoisonooo I'd recommend doing a hf login with a key that has access to the repo (it's easy to get access)

[tpoisonooo]

• Meta reject the download request
• and CN blocked HF and GitHub

What ancestral grave did we open-source developers dig up to deserve this .. ?

[terrykong]

@tpoisonooo Sorry you're running into this. I didn't realize they actually rejected anyone...

I'll PR something today that'll filter out the gated repos so you can at least run the tests that don't require those models.

[SahilJain314]

@tpoisonooo Updating here, PR #755 is merged. You should be able to run tests without HF with this.

[tpoisonooo]

@SahilJain314 @terrykong unittest passed.

[terrykong]

awesome work @tpoisonooo ! thanks for adding support, i've left some comments

[tpoisonooo]

awesome work @tpoisonooo ! thanks for adding support, i've left some comments

done.

[tpoisonooo]

@terrykong lint error fixed.

[terrykong]

great work! thanks for the contribution!

[terrykong]

@tpoisonooo looks like the linter failed. can you run it through pre-commit and make sure everything passes and we can try again?

[tpoisonooo]

For this PR

ok, I have installed pre-commit and passed lint check part, pls try again. @terrykong

pyrefly failed on other irrevalant files (convert_dcp_to_hf.py)

For code lint

I have seen a better practice: github actions bot merge format-code to human PR.

Here is an example: Tencent/ncnn@a2e23d9 in Tencent/ncnn#4782

[tpoisonooo]

my local pre-commit allows 2 empty lines here (that is why I dislike pre-commit), my python version is 3.10

[terrykong]

@tpoisonooo could you install pre-commit hooks like this?

uv run --group dev pre-commit install

if it doesn't work for you, i will help shepherd this PR in

[tpoisonooo]

@tpoisonooo could you install pre-commit hooks like this?

uv run --group dev pre-commit install

if it doesn't work for you, i will help shepherd this PR in

My mistake, python3 -m pip install pre-commit does not work for 2 empty lines.

Since I only changed 2 .py files and already manully fixed the lint error, nothing can be committed, please retry the CI. @terrykong

[tpoisonooo]

finally.. the CI passed.. QvQ

[Zeyi-Lin]

finally.. the CI passed.. QvQ

It seems not to have passed the Docs-Test QAQ, https://github.com/NVIDIA-NeMo/RL/actions/runs/16798324457/job/47576893936

[tpoisonooo]

finally.. the CI passed.. QvQ

It seems not to have passed the Docs-Test QAQ, https://github.com/NVIDIA-NeMo/RL/actions/runs/16798324457/job/47576893936

This looks like some errors in CI https://github.com/NVIDIA-NeMo/RL/actions/runs/16798324457/job/47576893936

• ModuleNotFoundError: No module named 'swanlab' , I already add dependency in pyproject.yaml
• NameError: name 'flatten_dict' is not defined

[terrykong]

hey @tpoisonooo . I'll take a look at this later today and help you get your PR in (i'll make sure to credit you for the contribution)

[terrykong]

closing since this is replaced with #923