Is there a way to disable data sampling?

[haozhouamzn]

Is your feature request related to a problem? Please describe.
Looks like with multiple datasets, e.g. in local_setup.yml, each dataset will be sampled certain % of data. For example, with two datasets, and each has 100 samples with weights [1,2]. Then it ends up with 33 samples from dataset A, and 66 samples from dataset B.

Is there a way to keep 100 samples of dataset A and 100 samples of dataset B?

Describe the solution you'd like
A flag or instruction on which codes need to be changed.

Describe alternatives you've considered
Overwrite the ratio in helper.cpp.

Additional context
I have a very large dataset, and I have to partition it into multiple small ones in order to process to mmap files. I want to train the model with all data.

[StellaAthena]

Is there a reason why using weights of [1, 1] doesn't accomplish what you want?