Enable mlperf to use mixtral-v1 tokenized dataset to avoid dropping in total_weights when splitting data across multiple host

[aireenmei]

Description

We observed drop in total_weights in multihost training. It comes from splitting untokenized data across host. Switch to tokenized dataset fixes it
b/394692604

Tests

self tested

Checklist

Before submitting this PR, please make sure (put X in square brackets):

• I have performed a self-review of my code.
• I have necessary comments in my code, particularly in hard-to-understand areas.
• I have run end-to-end tests tests and provided workload links above if applicable.
• I have made or will make corresponding changes to the doc if needed.