Update README.md

README.md

@@ -215,11 +215,6 @@ print(model.compute_score(sentence_pairs,
 
 
 We compare BGE-M3 with some popular methods, including BM25, openAI embedding, etc.
-We utilized Pyserini to implement BM25, and the test results can be reproduced by this [script](https://github.com/FlagOpen/FlagEmbedding/tree/master/C_MTEB/MLDR#bm25-baseline). 
-To make the BM25 and BGE-M3 more comparable, in the experiment, 
-BM25 used the same tokenizer as BGE-M3 (i.e., the tokenizer of XLM-Roberta). 
-Using the same vocabulary can also ensure that both approaches have the same retrieval latency. 
-
 
 - Multilingual (Miracl dataset) 
 
@@ -242,6 +237,12 @@ Using the same vocabulary can also ensure that both approaches have the same ret
   - NarritiveQA:  
   ![avatar](./imgs/nqa.jpg)
 
+- BM25  
+
+We utilized Pyserini to implement BM25, and the test results can be reproduced by this [script](https://github.com/FlagOpen/FlagEmbedding/tree/master/C_MTEB/MLDR#bm25-baseline).
+
+![avatar](./imgs/bm25.jpg)
+
 
 ## Training
 - Self-knowledge Distillation: combining multiple outputs from different 
@@ -259,7 +260,7 @@ Refer to our [report](https://arxiv.org/pdf/2402.03216.pdf) for more details.
 ## Acknowledgement
 
 Thanks the authors of open-sourced datasets, including Miracl, MKQA, NarritiveQA, etc. 
-Thanks the open-sourced libraries like [Tevatron](https://github.com/texttron/tevatron), [pyserial](https://github.com/pyserial/pyserial).
+Thanks the open-sourced libraries like [Tevatron](https://github.com/texttron/tevatron), [Pyserini](https://github.com/castorini/pyserini).