hparams are not logged in tensorboard

[pvnieo]

🐛 Bug

I'm using the latest version of PL. My problem is when I launch an experiment, everything is logged correctly to tensorboard, except the hparams:

I tried to launch the cpu_template on a fresh virtual environement, but it didn't work. However, when I launch the pytorch example of logging hparams to tensorboard (https://pytorch.org/docs/stable/tensorboard.html#torch.utils.tensorboard.writer.SummaryWriter.add_hparams), it worked!

To Reproduce

I succeeded to reproduce this using the official MNIST collab (https://colab.research.google.com/drive/1F_RNcHzTfFuQf-LeKvSlud6x7jXYkG31#scrollTo=HOk9c4_35FKg), by skipping the "Simplest example" and excuting directly '1. LightningModule'

Environment

• PyTorch Version (e.g., 1.0): 1.5
• OS (e.g., Linux): ubuntu 20.04
• How you installed PyTorch (conda, pip, source): pip
• Build command you used (if compiling from source):
• Python version: 3.8 / 3.6
• CUDA/cuDNN version: 10.2
• GPU models and configuration: GeForce GTX 1060
• Any other relevant information:

[david-waterworth]

Not sure if it's related - I'm getting hparams but not metrics #2406

[pvnieo]

UPDATE: I used the same code, with the same version in new folder (just copy past the code), and now I have the hparams, but no metrics!

[awaelchli]

putting a reminder here that we saw this problem popping up recently in one of the tests.
See here, #2512 (comment)
need to look at the test there to debug the problem.

[edenlightning]

Closing issue. @pvnieo please let us know if you have any issues!