ldwang commited on
Commit
1ad17b9
β€’
1 Parent(s): 60e9d3d
Files changed (1) hide show
  1. README.md +7 -6
README.md CHANGED
@@ -9,6 +9,7 @@ license: mit
9
  <h4 align="center">
10
  <p>
11
  <a href=#model-list>Model List</a> |
 
12
  <a href=#usage>Usage</a> |
13
  <a href="#evaluation">Evaluation</a> |
14
  <a href="#train">Train</a> |
@@ -28,8 +29,8 @@ And it also can be used in vector databases for LLMs.
28
 
29
  ************* 🌟**Updates**🌟 *************
30
  - 09/12/2023: New Release:
31
- - **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.
32
- - **update embedding model**: release bge-*-v1.5 embedding model to alleviate the issue of the similarity distribution, and enhance its retrieval ability without instruction.
33
  - 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.
34
  - 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).
35
  - 08/05/2023: Release base-scale and small-scale models, **best performance among the models of the same size πŸ€—**
@@ -61,7 +62,7 @@ And it also can be used in vector databases for LLMs.
61
 
62
  \*: 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.
63
 
64
- \**: To balance the accuracy and time cost, cross-encoder is widely used to re-rank top-k documents retrieved by other simple models.
65
  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.
66
 
67
 
@@ -73,7 +74,7 @@ For examples, use bge embedding model to retrieve top 100 relevant documents, an
73
  <!-- ### How to fine-tune bge embedding model? -->
74
  Following this [example](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune) to prepare data and fine-tune your model.
75
  Some suggestions:
76
- - Mine hard negatives following this [example](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune#data-format), which can improve the retrieval performance.
77
  - 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.
78
  - 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.
79
 
@@ -357,8 +358,8 @@ Cross-encoder will perform full-attention over the input pair,
357
  which is more accurate than embedding model (i.e., bi-encoder) but more time-consuming than embedding model.
358
  Therefore, it can be used to re-rank the top-k documents returned by embedding model.
359
  We train the cross-encoder on a multilingual pair data,
360
- The data format is the same as embedding model, so you can fine-tune it easily following our example.
361
- More details pelease refer to [./FlagEmbedding/reranker/README.md](./FlagEmbedding/reranker/README.md)
362
 
363
 
364
  ## Contact
 
9
  <h4 align="center">
10
  <p>
11
  <a href=#model-list>Model List</a> |
12
+ <a href=#frequently-asked-questions>FAQ</a> |
13
  <a href=#usage>Usage</a> |
14
  <a href="#evaluation">Evaluation</a> |
15
  <a href="#train">Train</a> |
 
29
 
30
  ************* 🌟**Updates**🌟 *************
31
  - 09/12/2023: New Release:
32
+ - **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.
33
+ - **update embedding model**: release `bge-*-v1.5` embedding model to alleviate the issue of the similarity distribution, and enhance its retrieval ability without instruction.
34
  - 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.
35
  - 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).
36
  - 08/05/2023: Release base-scale and small-scale models, **best performance among the models of the same size πŸ€—**
 
62
 
63
  \*: 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.
64
 
65
+ \**: 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.
66
  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.
67
 
68
 
 
74
  <!-- ### How to fine-tune bge embedding model? -->
75
  Following this [example](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune) to prepare data and fine-tune your model.
76
  Some suggestions:
77
+ - Mine hard negatives following this [example](https://github.com/FlagOpen/FlagEmbedding/tree/master/examples/finetune#hard-negatives), which can improve the retrieval performance.
78
  - 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.
79
  - 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.
80
 
 
358
  which is more accurate than embedding model (i.e., bi-encoder) but more time-consuming than embedding model.
359
  Therefore, it can be used to re-rank the top-k documents returned by embedding model.
360
  We train the cross-encoder on a multilingual pair data,
361
+ 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).
362
+ More details pelease refer to [./FlagEmbedding/reranker/README.md](https://github.com/FlagOpen/FlagEmbedding/tree/master/FlagEmbedding/reranker)
363
 
364
 
365
  ## Contact