[4/9] rm run_name, add_seed_and_date_to_exp_name

[garrett361]

Two related change:

• Remove the run_name arg in FlatArguments, because the user-supplied value is currently totally ignored and instead overwritten with data from the user-supplied exp_name arg.
• Add a new --add_seed_and_date_to_exp_name flag.

In upstream, the seed and runtime date are added to the wandb name and (unless --do_not_randomize_output_dir True is specified) the save path. This is hardcoded behavior.

This PR allows for skipping this addition by --add_seed_and_date_to_run_name False

This ability is important in (at least) the following two cases:

• When restarting a job, the run name would change due to the new run's updated time.
• In multi-node scenarios, different nodes would generically get different run name, due to different start times.

	PR	Title
1	#15	padding-free
2	#16	clean_checkpoints_at_end
3	#17	final_lr_ratio
4	>18	add_seed_and_date_to_run_name
5	#19	additional_model_arguments
6	#20	sync_each_batch=True grad acc
7	#21	no grad acc averaging for sum losses
8	#22	extra reporting
9	#23	local_main_process_first when building dataset

[fabianlim]

Im not sure if its a good idea to remove run_name

[garrett361]

Ok, I can undo it. Removing run_name doesn't actually change any behavior though, because run_name was never actually used. The user supplied value is just ignored :/

[garrett361]

It's true that removing run_name would require users to remove the arg in their configs, even though it is a no-op arg.

[garrett361]

So, remove?

[fabianlim]

oh i didnt know it was a no op.. ok fine lets remove

[fabianlim]

LGTM