Update README.md
#3
by
zhichao-geng
- opened
README.md
CHANGED
@@ -12,11 +12,6 @@ tags:
|
|
12 |
---
|
13 |
|
14 |
# opensearch-neural-sparse-encoding-doc-v1
|
15 |
-
This is a learned sparse retrieval model. It encodes the documents to 30522 dimensional **sparse vectors**. For queries, it just use a tokenizer and a weight look-up table to generate sparse vectors. The non-zero dimension index means the corresponding token in the vocabulary, and the weight means the importance of the token. And the similarity score is the inner product of query/document sparse vectors. In the real-world use case, the search performance of opensearch-neural-sparse-encoding-v1 is comparable to BM25.
|
16 |
-
|
17 |
-
This model is trained on MS MARCO dataset.
|
18 |
-
|
19 |
-
OpenSearch neural sparse feature supports learned sparse retrieval with lucene inverted index. Link: https://opensearch.org/docs/latest/query-dsl/specialized/neural-sparse/. The indexing and search can be performed with OpenSearch high-level API.
|
20 |
|
21 |
## Select the model
|
22 |
The model should be selected considering search relevance, model inference and retrieval efficiency(FLOPS). We benchmark models' **zero-shot performance** on a subset of BEIR benchmark: TrecCovid,NFCorpus,NQ,HotpotQA,FiQA,ArguAna,Touche,DBPedia,SCIDOCS,FEVER,Climate FEVER,SciFact,Quora.
|
@@ -31,6 +26,13 @@ Overall, the v2 series of models have better search relevance, efficiency and in
|
|
31 |
| [opensearch-neural-sparse-encoding-doc-v2-distill](https://huggingface.co/opensearch-project/opensearch-neural-sparse-encoding-doc-v2-distill) | ✔️ | 67M | 0.504 | 1.8 |
|
32 |
| [opensearch-neural-sparse-encoding-doc-v2-mini](https://huggingface.co/opensearch-project/opensearch-neural-sparse-encoding-doc-v2-mini) | ✔️ | 23M | 0.497 | 1.7 |
|
33 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
34 |
## Usage (HuggingFace)
|
35 |
This model is supposed to run inside OpenSearch cluster. But you can also use it outside the cluster, with HuggingFace models API.
|
36 |
|
|
|
12 |
---
|
13 |
|
14 |
# opensearch-neural-sparse-encoding-doc-v1
|
|
|
|
|
|
|
|
|
|
|
15 |
|
16 |
## Select the model
|
17 |
The model should be selected considering search relevance, model inference and retrieval efficiency(FLOPS). We benchmark models' **zero-shot performance** on a subset of BEIR benchmark: TrecCovid,NFCorpus,NQ,HotpotQA,FiQA,ArguAna,Touche,DBPedia,SCIDOCS,FEVER,Climate FEVER,SciFact,Quora.
|
|
|
26 |
| [opensearch-neural-sparse-encoding-doc-v2-distill](https://huggingface.co/opensearch-project/opensearch-neural-sparse-encoding-doc-v2-distill) | ✔️ | 67M | 0.504 | 1.8 |
|
27 |
| [opensearch-neural-sparse-encoding-doc-v2-mini](https://huggingface.co/opensearch-project/opensearch-neural-sparse-encoding-doc-v2-mini) | ✔️ | 23M | 0.497 | 1.7 |
|
28 |
|
29 |
+
## Overview
|
30 |
+
This is a learned sparse retrieval model. It encodes the documents to 30522 dimensional **sparse vectors**. For queries, it just use a tokenizer and a weight look-up table to generate sparse vectors. The non-zero dimension index means the corresponding token in the vocabulary, and the weight means the importance of the token. And the similarity score is the inner product of query/document sparse vectors. In the real-world use case, the search performance of opensearch-neural-sparse-encoding-v1 is comparable to BM25.
|
31 |
+
|
32 |
+
This model is trained on MS MARCO dataset.
|
33 |
+
|
34 |
+
OpenSearch neural sparse feature supports learned sparse retrieval with lucene inverted index. Link: https://opensearch.org/docs/latest/query-dsl/specialized/neural-sparse/. The indexing and search can be performed with OpenSearch high-level API.
|
35 |
+
|
36 |
## Usage (HuggingFace)
|
37 |
This model is supposed to run inside OpenSearch cluster. But you can also use it outside the cluster, with HuggingFace models API.
|
38 |
|