README.md
| 1 | --- |
| 2 | license: apache-2.0 |
| 3 | datasets: |
| 4 | - sentence-transformers/msmarco |
| 5 | language: |
| 6 | - en |
| 7 | base_model: |
| 8 | - cross-encoder/ms-marco-MiniLM-L12-v2 |
| 9 | pipeline_tag: text-ranking |
| 10 | library_name: sentence-transformers |
| 11 | tags: |
| 12 | - transformers |
| 13 | --- |
| 14 | # Cross-Encoder for MS Marco |
| 15 | |
| 16 | This model was trained on the [MS Marco Passage Ranking](https://github.com/microsoft/MSMARCO-Passage-Ranking) task. |
| 17 | |
| 18 | The model can be used for Information Retrieval: Given a query, encode the query will all possible passages (e.g. retrieved with ElasticSearch). Then sort the passages in a decreasing order. See [SBERT.net Retrieve & Re-rank](https://www.sbert.net/examples/applications/retrieve_rerank/README.html) for more details. The training code is available here: [SBERT.net Training MS Marco](https://github.com/UKPLab/sentence-transformers/tree/master/examples/cross_encoder/training/ms_marco) |
| 19 | |
| 20 | |
| 21 | ## Usage with SentenceTransformers |
| 22 | |
| 23 | The usage is easy when you have [SentenceTransformers](https://www.sbert.net/) installed. Then you can use the pre-trained models like this: |
| 24 | ```python |
| 25 | from sentence_transformers import CrossEncoder |
| 26 | |
| 27 | model = CrossEncoder('cross-encoder/ms-marco-MiniLM-L6-v2') |
| 28 | scores = model.predict([ |
| 29 | ("How many people live in Berlin?", "Berlin had a population of 3,520,031 registered inhabitants in an area of 891.82 square kilometers."), |
| 30 | ("How many people live in Berlin?", "Berlin is well known for its museums."), |
| 31 | ]) |
| 32 | print(scores) |
| 33 | # [ 8.607138 -4.320078] |
| 34 | ``` |
| 35 | |
| 36 | |
| 37 | ## Usage with Transformers |
| 38 | |
| 39 | ```python |
| 40 | from transformers import AutoTokenizer, AutoModelForSequenceClassification |
| 41 | import torch |
| 42 | |
| 43 | model = AutoModelForSequenceClassification.from_pretrained('cross-encoder/ms-marco-MiniLM-L6-v2') |
| 44 | tokenizer = AutoTokenizer.from_pretrained('cross-encoder/ms-marco-MiniLM-L6-v2') |
| 45 | |
| 46 | features = tokenizer(['How many people live in Berlin?', 'How many people live in Berlin?'], ['Berlin has a population of 3,520,031 registered inhabitants in an area of 891.82 square kilometers.', 'New York City is famous for the Metropolitan Museum of Art.'], padding=True, truncation=True, return_tensors="pt") |
| 47 | |
| 48 | model.eval() |
| 49 | with torch.no_grad(): |
| 50 | scores = model(**features).logits |
| 51 | print(scores) |
| 52 | ``` |
| 53 | |
| 54 | |
| 55 | ## Performance |
| 56 | In the following table, we provide various pre-trained Cross-Encoders together with their performance on the [TREC Deep Learning 2019](https://microsoft.github.io/TREC-2019-Deep-Learning/) and the [MS Marco Passage Reranking](https://github.com/microsoft/MSMARCO-Passage-Ranking/) dataset. |
| 57 | |
| 58 | |
| 59 | | Model-Name | NDCG@10 (TREC DL 19) | MRR@10 (MS Marco Dev) | Docs / Sec | |
| 60 | | ------------- |:-------------| -----| --- | |
| 61 | | **Version 2 models** | | | |
| 62 | | cross-encoder/ms-marco-TinyBERT-L2-v2 | 69.84 | 32.56 | 9000 |
| 63 | | cross-encoder/ms-marco-MiniLM-L2-v2 | 71.01 | 34.85 | 4100 |
| 64 | | cross-encoder/ms-marco-MiniLM-L4-v2 | 73.04 | 37.70 | 2500 |
| 65 | | cross-encoder/ms-marco-MiniLM-L6-v2 | 74.30 | 39.01 | 1800 |
| 66 | | cross-encoder/ms-marco-MiniLM-L12-v2 | 74.31 | 39.02 | 960 |
| 67 | | **Version 1 models** | | | |
| 68 | | cross-encoder/ms-marco-TinyBERT-L2 | 67.43 | 30.15 | 9000 |
| 69 | | cross-encoder/ms-marco-TinyBERT-L4 | 68.09 | 34.50 | 2900 |
| 70 | | cross-encoder/ms-marco-TinyBERT-L6 | 69.57 | 36.13 | 680 |
| 71 | | cross-encoder/ms-marco-electra-base | 71.99 | 36.41 | 340 |
| 72 | | **Other models** | | | |
| 73 | | nboost/pt-tinybert-msmarco | 63.63 | 28.80 | 2900 |
| 74 | | nboost/pt-bert-base-uncased-msmarco | 70.94 | 34.75 | 340 |
| 75 | | nboost/pt-bert-large-msmarco | 73.36 | 36.48 | 100 |
| 76 | | Capreolus/electra-base-msmarco | 71.23 | 36.89 | 340 |
| 77 | | amberoad/bert-multilingual-passage-reranking-msmarco | 68.40 | 35.54 | 330 |
| 78 | | sebastian-hofstaetter/distilbert-cat-margin_mse-T2-msmarco | 72.82 | 37.88 | 720 |
| 79 | |
| 80 | Note: Runtime was computed on a V100 GPU. |