Update readme

README.md

@@ -1,25 +1,26 @@
-# rdr-question_encoder-single-trivia-base
+# rdr-queston_encoder-single-nq-base
 
 Reader-Distilled Retriever (`RDR`)
 
 Sohee Yang and Minjoon Seo, [Is Retriever Merely an Approximator of Reader?](https://arxiv.org/abs/2010.10999), arXiv 2020
 
-The paper proposes to distill the reader into the retriever so that the retriever absorbs the strength of the reader while keeping its own benefit. The model is a DPR retriever further finetuned using knowledge distillation from the DPR reader. Using this approach, the answer recall rate increases by a large margin, especially at a small number of top-k.
+The paper proposes to distill the reader into the retriever so that the retriever absorbs the strength of the reader while keeping its own benefit. The model is a DPR retriever further finetuned using knowledge distillation from the DPR reader. Using this approach, the answer recall rate increases by a large margin, especially at small numbers of top-k.
 
-This model is the question encoder of RDR trained solely on TriviaQA (single-trivia). This model is trained by the authors and is the official checkpoint of RDR.
+This model is the question encoder of RDR trained solely on Natural Questions (NQ) (single-nq). This model is trained by the authors and is the official checkpoint of RDR.
 
 ## Performance
 
 The following is the answer recall rate measured using PyTorch 1.4.0 and transformers 4.5.0.
 
-For the values of DPR, those in parentheses are directly taken from the paper. The values without parentheses are reported using the reproduction of DPR that consists of [this question encoder](https://huggingface.co/soheeyang/dpr-question_encoder-single-trivia-base) and [this queston encoder](https://huggingface.co/soheeyang/dpr-question_encoder-single-trivia-base).
+The values of DPR on the NQ dev set are taken from Table 1 of the paper of [RDR](https://arxiv.org/abs/2010.10999). The values of DPR on the NQ test set are taken from the [codebase of DPR](https://github.com/facebookresearch/DPR). DPR-adv is the a new DPR model released in March 2021. It is trained on the original DPR NQ train set and its version where hard negatives are mined using DPR index itself using the previous NQ checkpoint. Please refer to the [codebase of DPR](https://github.com/facebookresearch/DPR) for more details about DPR-adv-hn.
 
-|             | Top-K Passages   | 1         | 5         | 20        | 50        | 100       |
-|-------------|------------------|-----------|-----------|-----------|-----------|-----------|
-|**TriviaQA Dev** | **DPR**              | 54.27     | 71.11     | 79.53     | 82.72     | 85.07     |
-|             | **RDR (This Model)** | **61.84** | **75.93** | **82.56** | **85.35** | **87.00** |
-|**TriviaQA Test**| **DPR**              | 54.41     | 70.99     | 79.31 (79.4)     | 82.90     | 84.99 (85.0)     |
-|             | **RDR (This Model)** | **62.56** | **75.92** | **82.52** | **85.64** | **87.26** |
+|         | Top-K Passages   | 1     | 5     | 20    | 50    | 100   |
+|---------|------------------|-------|-------|-------|-------|-------|
+| **NQ Dev**  | **DPR (approx)** | 44.2 | - | 76.9 | 81.3  | 84.2 |
+|             | **RDR (This Model)** | **54.43** | **72.17** | **81.33** | **84.8**  | **86.61** |
+| **NQ Test** | **DPR**              | 45.87 | 68.14 | 79.97 | -     | 85.87 |
+|         | **DPR-adv-hn**          | 52.47 | **72.24** | 81.33 | -     | 87.29 |
+|         | **RDR (This Model)** | **54.29** | 72.16 | **82.8**  | **86.34** | **88.2**  |
 
 ## How to Use
 
@@ -32,8 +33,8 @@ Therefore, please specify the exact class to use the model.
 ```python
 from transformers import DPRQuestionEncoder, AutoTokenizer
 
-tokenizer = AutoTokenizer.from_pretrained("soheeyang/rdr-question_encoder-single-trivia-base")
-question_encoder = DPRQuestionEncoder.from_pretrained("soheeyang/rdr-question_encoder-single-trivia-base")
+tokenizer = AutoTokenizer.from_pretrained("soheeyang/rdr-question_encoder-single-nq-base")
+question_encoder = DPRQuestionEncoder.from_pretrained("soheeyang/rdr-question_encoder-single-nq-base")
 
 data = tokenizer("question comes here", return_tensors="pt")
 question_embedding = question_encoder(**data).pooler_output  # embedding vector for question