ldwang commited on
Commit
9d3e225
β€’
1 Parent(s): bc5e88a
Files changed (1) hide show
  1. README.md +7 -6
README.md CHANGED
@@ -12,6 +12,7 @@ pipeline_tag: sentence-similarity
12
  <h4 align="center">
13
  <p>
14
  <a href=#model-list>Model List</a> |
 
15
  <a href=#usage>Usage</a> |
16
  <a href="#evaluation">Evaluation</a> |
17
  <a href="#train">Train</a> |
@@ -31,8 +32,8 @@ And it also can be used in vector databases for LLMs.
31
 
32
  ************* 🌟**Updates**🌟 *************
33
  - 09/12/2023: New Release:
34
- - **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.
35
- - **update embedding model**: release bge-*-v1.5 embedding model to alleviate the issue of the similarity distribution, and enhance its retrieval ability without instruction.
36
  - 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.
37
  - 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).
38
  - 08/05/2023: Release base-scale and small-scale models, **best performance among the models of the same size πŸ€—**
@@ -64,7 +65,7 @@ And it also can be used in vector databases for LLMs.
64
 
65
  \*: 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.
66
 
67
- \**: To balance the accuracy and time cost, cross-encoder is widely used to re-rank top-k documents retrieved by other simple models.
68
  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.
69
 
70
 
@@ -76,7 +77,7 @@ For examples, use bge embedding model to retrieve top 100 relevant documents, an
76
  <!-- ### How to fine-tune bge embedding model? -->
77
  Following this [example](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune) to prepare data and fine-tune your model.
78
  Some suggestions:
79
- - Mine hard negatives following this [example](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune#data-format), which can improve the retrieval performance.
80
  - 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.
81
  - 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.
82
 
@@ -360,8 +361,8 @@ Cross-encoder will perform full-attention over the input pair,
360
  which is more accurate than embedding model (i.e., bi-encoder) but more time-consuming than embedding model.
361
  Therefore, it can be used to re-rank the top-k documents returned by embedding model.
362
  We train the cross-encoder on a multilingual pair data,
363
- The data format is the same as embedding model, so you can fine-tune it easily following our example.
364
- More details pelease refer to [./FlagEmbedding/reranker/README.md](./FlagEmbedding/reranker/README.md)
365
 
366
 
367
  ## Contact
 
12
  <h4 align="center">
13
  <p>
14
  <a href=#model-list>Model List</a> |
15
+ <a href=#frequently-asked-questions>FAQ</a> |
16
  <a href=#usage>Usage</a> |
17
  <a href="#evaluation">Evaluation</a> |
18
  <a href="#train">Train</a> |
 
32
 
33
  ************* 🌟**Updates**🌟 *************
34
  - 09/12/2023: New Release:
35
+ - **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.
36
+ - **update embedding model**: release `bge-*-v1.5` embedding model to alleviate the issue of the similarity distribution, and enhance its retrieval ability without instruction.
37
  - 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.
38
  - 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).
39
  - 08/05/2023: Release base-scale and small-scale models, **best performance among the models of the same size πŸ€—**
 
65
 
66
  \*: 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.
67
 
68
+ \**: 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.
69
  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.
70
 
71
 
 
77
  <!-- ### How to fine-tune bge embedding model? -->
78
  Following this [example](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune) to prepare data and fine-tune your model.
79
  Some suggestions:
80
+ - Mine hard negatives following this [example](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune#hard-negatives), which can improve the retrieval performance.
81
  - 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.
82
  - 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.
83
 
 
361
  which is more accurate than embedding model (i.e., bi-encoder) but more time-consuming than embedding model.
362
  Therefore, it can be used to re-rank the top-k documents returned by embedding model.
363
  We train the cross-encoder on a multilingual pair data,
364
+ 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).
365
+ More details pelease refer to [./FlagEmbedding/reranker/README.md](https://github.com/FlagOpen/FlagEmbedding/tree/master/FlagEmbedding/reranker)
366
 
367
 
368
  ## Contact