Trainer + wandb quality of life logging tweaks

[TevenLeScao]

As discussed on Slack, this PR adds the possibility for users to specify a name for the run in their wandb project, and logs the model config in addition to the trainer args in the wandb.init call (if there is a duplicate key, it is overriden by the trainer args)

[codecov]

Codecov Report

Merging #6241 into master will decrease coverage by 0.03%.
The diff coverage is 55.55%.

@@            Coverage Diff             @@
##           master    #6241      +/-   ##
==========================================
- Coverage   79.61%   79.57%   -0.04%     
==========================================
  Files         146      146              
  Lines       26597    26600       +3     
==========================================
- Hits        21175    21167       -8     
- Misses       5422     5433      +11     

Impacted Files	Coverage Δ
src/transformers/configuration_encoder_decoder.py	100.00% <ø> (ø)
src/transformers/modeling_encoder_decoder.py	92.20% <ø> (ø)
src/transformers/trainer.py	39.06% <0.00%> (-0.09%)	⬇️
src/transformers/trainer_tf.py	12.45% <0.00%> (-0.05%)	⬇️
src/transformers/training_args_tf.py	47.45% <ø> (ø)
src/transformers/benchmark/benchmark_args_utils.py	89.13% <100.00%> (ø)
src/transformers/tokenization_utils_base.py	93.84% <100.00%> (ø)
src/transformers/training_args.py	80.39% <100.00%> (+0.19%)	⬆️
src/transformers/generation_tf_utils.py	79.19% <0.00%> (-2.51%)	⬇️
src/transformers/file_utils.py	80.30% <0.00%> (+0.25%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d5b0a0e...f588bd6. Read the comment docs.

[sgugger]

Thanks for doing this! Could you copy the lines in setup for TFTrainer as well and the doc line in TFTrainingArguments? (no need to redefine it there with field as TFTrainingArguments subclasses TFTrainingArguments).

[LysandreJik]

Looks good!

[TevenLeScao]

Updated the PR with your comments!

[TevenLeScao]

But CI looks like it broke inbetween

[borisdayma]

I love the fact that more parameters are now auto logged.

For info the approach right now in this integration was to pass environment variables to configure wandb.init so you can set many more parameters: https://docs.wandb.com/library/environment-variables

You can also call wandb.init yourself before which will replace any future ones with a noop.

People use a lot of different init parameters so you can either add them manually (like with the run_name one) or could use a "kwargs" approach to accept more (and extract all the valid ones or even just the ones starting with wandb_).

[dsblank]

Hello @LysandreJik and friends! If you like more items being logged, hopefully someone can review #6176 sooner than later to cut down on merge conflicts. I re-arranged some of the wandb and tensorboard initialization code to better accommodate what could be a growing list of integrations.

[TevenLeScao]

@LysandreJik @dsblank do you think we should merge #6176 first?

[sgugger]

Merging this right now since it's ready. @dsblank Sorry for the delay in reviewing your PR, I'm heading there next and will help solve potential merge conflicts if needed.