ldwang commited on
Commit
75d0064
β€’
1 Parent(s): 0725d0e
Files changed (1) hide show
  1. README.md +9 -6
README.md CHANGED
@@ -2603,12 +2603,15 @@ pipeline_tag: sentence-similarity
2603
  ---
2604
 
2605
 
 
 
2606
  <h1 align="center">FlagEmbedding</h1>
2607
 
2608
 
2609
  <h4 align="center">
2610
  <p>
2611
  <a href=#model-list>Model List</a> |
 
2612
  <a href=#usage>Usage</a> |
2613
  <a href="#evaluation">Evaluation</a> |
2614
  <a href="#train">Train</a> |
@@ -2628,8 +2631,8 @@ And it also can be used in vector databases for LLMs.
2628
 
2629
  ************* 🌟**Updates**🌟 *************
2630
  - 09/12/2023: New Release:
2631
- - **New reranker model**: release a cross-encoder model bge-reranker-base, which is more powerful than embedding model. We recommend to use/fine-tune it to re-rank top-k documents returned by embedding models.
2632
- - **update embedding model**: release bge-*-v1.5 embedding model to alleviate the issue of the similarity distribution, and enhance its retrieval ability without instruction.
2633
  - 09/07/2023: Update [fine-tune code](https://github.com/FlagOpen/FlagEmbedding/blob/master/FlagEmbedding/baai_general_embedding/README.md): Add script to mine hard negatives and support adding instruction during fine-tuning.
2634
  - 08/09/2023: BGE Models are integrated into **Langchain**, you can use it like [this](#using-langchain); C-MTEB **leaderboard** is [available](https://huggingface.co/spaces/mteb/leaderboard).
2635
  - 08/05/2023: Release base-scale and small-scale models, **best performance among the models of the same size πŸ€—**
@@ -2661,7 +2664,7 @@ And it also can be used in vector databases for LLMs.
2661
 
2662
  \*: If you need to search the relevant passages to a query, we suggest to add the instruction to the query; in other cases, no instruction is needed, just use the original query directly. In all cases, **no instruction** needs to be added to passages.
2663
 
2664
- \**: To balance the accuracy and time cost, cross-encoder is widely used to re-rank top-k documents retrieved by other simple models.
2665
  For examples, use bge embedding model to retrieve top 100 relevant documents, and then use bge reranker to re-rank the top 100 document to get the final top-3 results.
2666
 
2667
 
@@ -2673,7 +2676,7 @@ For examples, use bge embedding model to retrieve top 100 relevant documents, an
2673
  <!-- ### How to fine-tune bge embedding model? -->
2674
  Following this [example](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune) to prepare data and fine-tune your model.
2675
  Some suggestions:
2676
- - Mine hard negatives following this [example](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune#data-format), which can improve the retrieval performance.
2677
  - If you pre-train bge on your data, the pre-trained model cannot be directly used to calculate similarity, and it must be fine-tuned with contrastive learning before computing similarity.
2678
  - If the accuracy of the fine-tuned model is still not high, it is recommended to use/fine-tune the cross-encoder model (bge-reranker) to re-rank top-k results. Hard negatives also are needed to fine-tune reranker.
2679
 
@@ -2957,8 +2960,8 @@ Cross-encoder will perform full-attention over the input pair,
2957
  which is more accurate than embedding model (i.e., bi-encoder) but more time-consuming than embedding model.
2958
  Therefore, it can be used to re-rank the top-k documents returned by embedding model.
2959
  We train the cross-encoder on a multilingual pair data,
2960
- The data format is the same as embedding model, so you can fine-tune it easily following our example.
2961
- More details pelease refer to [./FlagEmbedding/reranker/README.md](./FlagEmbedding/reranker/README.md)
2962
 
2963
 
2964
  ## Contact
 
2603
  ---
2604
 
2605
 
2606
+ **Recommend switching to newest bge-base-en-v1.5, which has more reasonable similarity distribution and same method of usage.**
2607
+
2608
  <h1 align="center">FlagEmbedding</h1>
2609
 
2610
 
2611
  <h4 align="center">
2612
  <p>
2613
  <a href=#model-list>Model List</a> |
2614
+ <a href=#frequently-asked-questions>FAQ</a> |
2615
  <a href=#usage>Usage</a> |
2616
  <a href="#evaluation">Evaluation</a> |
2617
  <a href="#train">Train</a> |
 
2631
 
2632
  ************* 🌟**Updates**🌟 *************
2633
  - 09/12/2023: New Release:
2634
+ - **New reranker model**: release cross-encoder models `BAAI/bge-reranker-base` and `BAAI/bge-reranker-large`, which are more powerful than embedding model. We recommend to use/fine-tune them to re-rank top-k documents returned by embedding models.
2635
+ - **update embedding model**: release `bge-*-v1.5` embedding model to alleviate the issue of the similarity distribution, and enhance its retrieval ability without instruction.
2636
  - 09/07/2023: Update [fine-tune code](https://github.com/FlagOpen/FlagEmbedding/blob/master/FlagEmbedding/baai_general_embedding/README.md): Add script to mine hard negatives and support adding instruction during fine-tuning.
2637
  - 08/09/2023: BGE Models are integrated into **Langchain**, you can use it like [this](#using-langchain); C-MTEB **leaderboard** is [available](https://huggingface.co/spaces/mteb/leaderboard).
2638
  - 08/05/2023: Release base-scale and small-scale models, **best performance among the models of the same size πŸ€—**
 
2664
 
2665
  \*: If you need to search the relevant passages to a query, we suggest to add the instruction to the query; in other cases, no instruction is needed, just use the original query directly. In all cases, **no instruction** needs to be added to passages.
2666
 
2667
+ \**: Different embedding model, reranker is a cross-encoder, which cannot be used to generate embedding. To balance the accuracy and time cost, cross-encoder is widely used to re-rank top-k documents retrieved by other simple models.
2668
  For examples, use bge embedding model to retrieve top 100 relevant documents, and then use bge reranker to re-rank the top 100 document to get the final top-3 results.
2669
 
2670
 
 
2676
  <!-- ### How to fine-tune bge embedding model? -->
2677
  Following this [example](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune) to prepare data and fine-tune your model.
2678
  Some suggestions:
2679
+ - Mine hard negatives following this [example](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune#hard-negatives), which can improve the retrieval performance.
2680
  - If you pre-train bge on your data, the pre-trained model cannot be directly used to calculate similarity, and it must be fine-tuned with contrastive learning before computing similarity.
2681
  - If the accuracy of the fine-tuned model is still not high, it is recommended to use/fine-tune the cross-encoder model (bge-reranker) to re-rank top-k results. Hard negatives also are needed to fine-tune reranker.
2682
 
 
2960
  which is more accurate than embedding model (i.e., bi-encoder) but more time-consuming than embedding model.
2961
  Therefore, it can be used to re-rank the top-k documents returned by embedding model.
2962
  We train the cross-encoder on a multilingual pair data,
2963
+ The data format is the same as embedding model, so you can fine-tune it easily following our [example](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/reranker).
2964
+ More details pelease refer to [./FlagEmbedding/reranker/README.md](https://github.com/FlagOpen/FlagEmbedding/tree/master/FlagEmbedding/reranker)
2965
 
2966
 
2967
  ## Contact