Datasets:

---

mteb

/

msmarco-v2

like 0

Follow

Massive Text Embedding Benchmark 526

Hello,

0.
rows count severe mismatch!

• queries: ~1M in microsoft/msmarco-v2 , 285k in mteb/msmarco-v2
• corpus: ~1M * ~10 passages per query in microsoft/msmarco-v2, 133M in mteb/msmarco-v2

1. There seem to be missing which passages were originally assigned to each query - in original ms marco, there is list of ~10 passages per query. The format here strange, multiple rows having same title etc. without explicit information whether the rows belong to each other.

2. In the original msmarco-v2, there is 'answer' column. Why it is missing in mteb version? Yo, where 'answer' at?

Thank you,

w/regards
D