Fix typo in Mixtral accuracy

[pgmpablo157321]

No description provided.

[github-actions]

MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅

[szutenberg]

The typo correction is fine but in these rules I see few discrepancies:

• OpenOrca has only train spit, there is no "GPT-4 split": The OpenOrca dataset is a collection of augmented FLAN Collection data. Currently ~1M GPT-4 completions, and ~3.2M GPT-3.5 completions.
• we used GSM8K samples from the train split (train has 7.47k, test 1.32k)
• MBXP has only test split - no need to specify
• no need to duplicate "5k samples, max_seq_len=2048". Consider writing it in a simpler form: "curated dataset containing 5k samples from GSM8K, OpenOrca, MBXP each, max_seq_len=2048
• tokens per sample should be 145.9
• there is only one accuracy category: 99%. Reference outputs are for fp16.
• rouge scores are for OpenOrca samples only

[mrmhodak]

We can merge during WG meeting 7/2.

@szutenberg : Can you create a PR for the discrepancies you found?

[mrmhodak]

@pgmpablo157321 to create PR