mC4 sampling & pre-processing

Hi @TevenLeScao,

I think there are some confusing and broken link in the [mC4 data preprocessing](https://github.com/bigscience-workshop/bigscience/tree/master/data/mc4) section. Can you take a look?

Both of the links are broken here,
1. [mc4_preprocessing](https://console.cloud.google.com/storage/browser/bigscience/mc4_preprocessing?pageState=(%22StorageObjectListTable%22:(%22f%22:%22%255B%255D%22)))
2. [mc4_sampled_raw](https://console.cloud.google.com/storage/browser/bigscience/mc4_sampled_raw?pageState=(%22StorageObjectListTable%22:(%22f%22:%22%255B%255D%22)))

The original link should be,

1. [mc4_preprocessing](https://console.cloud.google.com/storage/browser/bigscience/preprocessed_data/mc4_preprocessing;tab=objects?pageState=(%22StorageObjectListTable%22:(%22f%22:%22%255B%255D%22))&prefix=&forceOnObjectsSortingFiltering=false)
2. [mc4_sampled_raw](https://console.cloud.google.com/storage/browser/bigscience/preprocessed_data/mc4_sampled_raw?pageState=(%22StorageObjectListTable%22:(%22f%22:%22%255B%255D%22))&prefix=&forceOnObjectsSortingFiltering=false)

In addition to that, the multinomial data processing code to create the different language splits are in this pull request, https://github.com/bigscience-workshop/Megatron-DeepSpeed/pull/9

Here's few things,

1. Did you use this data for any one of your experiments?
2. If not then I think you can update the doc, https://github.com/bigscience-workshop/bigscience/tree/master/data/mc4

For reference purpose, if you want to keep the code, I'm happy to open a pull request here. If not I'll close the pull request from [bigscience/Megatron-Deepspeed](https://github.com/bigscience-workshop/Megatron-DeepSpeed/pull/9) repo.

Let me know what do you think.

Beginne mit der im Issue verlinkten Dokumentation zu data/mc4 und vergleiche die beiden mC4-Links zu Speicherorten mit ihren vorgesehenen Zielen. Überprüfe, dass die korrigierten Speicherorte weiterhin gültig sind, und kläre oder aktualisiere die Verweise auf das Preprocessing; abgeschlossen ist die Aufgabe, wenn die Dokumentationslinks funktionieren und der Umfang rund um den zugehörigen Verarbeitungscode festgelegt ist.

Vom Indexierungsmodell aus dem Issue-Text verfasst.