readme

README.md

@@ -27,7 +27,7 @@ You can stream the dataset for example like this:
 ```python
 from datasets import load_dataset
 
-dataset = load_dataset("Cohere/msmarco-v2.1-embed-english-v3", split="passages", streaming=True)
+dataset = load_dataset("Cohere/msmarco-v2.1-embed-english-v3", "passages", split="train", streaming=True)
 
 for row in dataset:
     print(row)
@@ -58,13 +58,21 @@ The format is the following:
 - "qrels": Relevance annotations for the 215 annotated queries by NIST
 
 ### JSONL
-The folder `queries_jsonl/` contains the queries, including the query id, the query text, the embedding,
+The folder `queries_jsonl/` contains the queries in a `.jsonl.gz` format.
+
+Load this .jsonl.gz format is significantly faster than the parquet file.
+
+
 ### Parquet
 
+If you want to use the parquet file or the HF datasets library, the folder `queries_parquet/` contains the respective parquet file.
+
+You can load the queries with the following command in HF datasets
+
 ```python
 from datasets import load_dataset
 
-dataset = load_dataset("Cohere/msmarco-v2.1-embed-english-v3", split="passages", streaming=True)
+dataset = load_dataset("Cohere/msmarco-v2.1-embed-english-v3", "queries", split="test", streaming=True)
 
 for row in dataset:
     print(row)