Llama-8B missing SingleStream accuracy

[nvzhihanj]

@pgmpablo157321 discovered that we didn't have SS dataset and accuracy for the 5000 sub-samples athttps://github.com/mlcommons/inference/tree/master/language/llama3.1-8b#5000-samples-edge .
@attafosu I believe Taran and Tianmu mentioned this will be released but we couldn't find them.

[anivar]

I can help with this missing SingleStream dataset issue. I've been working on preprocessing documentation improvements (PR #2270) and noticed similar gaps in dataset availability documentation.

For the 5000 samples edge case, it appears we need:

1. The specific SingleStream dataset split (5000 samples)
2. Corresponding accuracy metrics/ground truth
3. Clear documentation on where these files should be hosted

I'll investigate the dataset structure and work with the team to ensure these files are properly generated and documented. This aligns with my ongoing work to improve dataset preprocessing documentation across models.