Update README.md
Browse files
README.md
CHANGED
@@ -1,40 +1,43 @@
|
|
1 |
-
# MS MARCO Passages Hard Negatives
|
2 |
-
|
3 |
-
[
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
10 |
-
##
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
distilbert-
|
30 |
-
distilbert-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
distilbert-margin_mse-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
40 |
-
|
|
|
|
|
|
|
|
1 |
+
# MS MARCO Passages Hard Negatives
|
2 |
+
|
3 |
+
> [!NOTE]
|
4 |
+
> This repository contains raw datasets, all of which have also been formatted for easy training in the [MS MARCO Mined Triplets](https://huggingface.co/collections/sentence-transformers/ms-marco-mined-triplets-6644d6f1ff58c5103fe65f23) collection. We recommend looking there first.
|
5 |
+
|
6 |
+
[MS MARCO](https://microsoft.github.io/msmarco/) is a large scale information retrieval corpus that was created based on real user search queries using Bing search engine.
|
7 |
+
|
8 |
+
This dataset repository contains files that are helpful to train bi-encoder models e.g. using [sentence-transformers](https://www.sbert.net).
|
9 |
+
|
10 |
+
## Training Code
|
11 |
+
You can find here an example how these files can be used to train bi-encoders: [SBERT.net - MS MARCO - MarginMSE](https://www.sbert.net/examples/training/ms_marco/README.html#marginmse)
|
12 |
+
|
13 |
+
## cross-encoder-ms-marco-MiniLM-L-6-v2-scores.pkl.gz
|
14 |
+
|
15 |
+
This is a pickled dictionary in the format: `scores[qid][pid] -> cross_encoder_score`
|
16 |
+
|
17 |
+
It contains 160 million cross-encoder scores for (query, paragraph) pairs using the [cross-encoder/ms-marco-MiniLM-L-6-v2](https://huggingface.co/cross-encoder/ms-marco-MiniLM-L-6-v2) model.
|
18 |
+
|
19 |
+
## msmarco-hard-negatives.jsonl.gz
|
20 |
+
This is a jsonl file: Each line is a JSON object. It has the following format:
|
21 |
+
```
|
22 |
+
{"qid": 867436, "pos": [5238393], "neg": {"bm25": [...], ...}}
|
23 |
+
```
|
24 |
+
|
25 |
+
`qid` is the query-ID from MS MARCO, `pos` is a list with paragraph IDs for positive passages. `neg` is a dictionary where we mined hard negatives using different (mainly dense retrieval) systems.
|
26 |
+
|
27 |
+
It contains hard negatives mined from BM25 (using ElasticSearch) and the following dense models:
|
28 |
+
```
|
29 |
+
msmarco-distilbert-base-tas-b
|
30 |
+
msmarco-distilbert-base-v3
|
31 |
+
msmarco-MiniLM-L-6-v3
|
32 |
+
distilbert-margin_mse-cls-dot-v2
|
33 |
+
distilbert-margin_mse-cls-dot-v1
|
34 |
+
distilbert-margin_mse-mean-dot-v1
|
35 |
+
mpnet-margin_mse-mean-v1
|
36 |
+
co-condenser-margin_mse-cls-v1
|
37 |
+
distilbert-margin_mse-mnrl-mean-v1
|
38 |
+
distilbert-margin_mse-sym_mnrl-mean-v1
|
39 |
+
distilbert-margin_mse-sym_mnrl-mean-v2
|
40 |
+
co-condenser-margin_mse-sym_mnrl-mean-v1
|
41 |
+
```
|
42 |
+
|
43 |
+
From each system, 50 most similar paragraphs were mined for a given query.
|