Hugging Face's logo

KaLM-Embedding
/
KaLM-Reranker-V1-Nano-R2

Text Ranking
sentence-transformers
Safetensors
Transformers
multilingual
t5gemma2
text2text-generation
reranker
encoder-decoder
FBNL
matryoshka
retrieval
RAG
Model card Files
xet
 Community
Yuki131 commited on
Commit
688b4ab
·
verified ·
1 Parent(s): 116813b

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +5 -3
README.md CHANGED
@@ -51,8 +51,6 @@ library_name: sentence-transformers
51
 
52
  ![kalm-reranker-v1-r2 architecture](./assets/framework.jpg)
53
 
54
- ![kalm-reranker-v1-r2 training](./assets/training.jpg)
55
-
56
  | Capability | KaLM-Reranker-V1 | KaLM-Reranker-V1-R2 |
57
  |:--|:--|:--|
58
  | Validated MEP range | 1x-32x | **1x-128x** |
@@ -61,7 +59,6 @@ library_name: sentence-transformers
61
  | Multi-domain ranking | Strong | **Improved on BEIR for all three sizes** |
62
  | Multilingual ranking | Limited | **Substantially improved across MIRACL's 18 languages** |
63
 
64
-
65
  ### KaLM-Reranker-V1-R2 vs. the original KaLM-Reranker-V1
66
  All results below use the same default compression ratio, `r=4`. KaLM-Reranker-V1 results are taken from the previous model cards; KaLM-Reranker-V1-R2 results are from the updated paper.
67
 
@@ -155,6 +152,11 @@ KaLM-Reranker-V1 achieves competitive performance on MIRACL. Within its paramete
155
  ko, ru, and sw); six of these languages (i.e., en, es, de, ja, fr, and ru) have reported website usage rates totaling approximately 74.2% among websites with known content languages ([W3Techs, 2026](https://w3techs.com/technologies/overview/content_language)).
156
  ![miracl](./assets/miracl.jpg)
157
 
 
 
 
 
 
158
  ### Acknowledgements
159
  We sincerely thank `jina-reranker-v3` and `Qwen3-Reranker` for their valuable inspiration and contributions to the reranking community, from which we have learned a lot.
160
 
 
51
 
52
  ![kalm-reranker-v1-r2 architecture](./assets/framework.jpg)
53
 
 
 
54
  | Capability | KaLM-Reranker-V1 | KaLM-Reranker-V1-R2 |
55
  |:--|:--|:--|
56
  | Validated MEP range | 1x-32x | **1x-128x** |
 
59
  | Multi-domain ranking | Strong | **Improved on BEIR for all three sizes** |
60
  | Multilingual ranking | Limited | **Substantially improved across MIRACL's 18 languages** |
61
 
 
62
  ### KaLM-Reranker-V1-R2 vs. the original KaLM-Reranker-V1
63
  All results below use the same default compression ratio, `r=4`. KaLM-Reranker-V1 results are taken from the previous model cards; KaLM-Reranker-V1-R2 results are from the updated paper.
64
 
 
152
  ko, ru, and sw); six of these languages (i.e., en, es, de, ja, fr, and ru) have reported website usage rates totaling approximately 74.2% among websites with known content languages ([W3Techs, 2026](https://w3techs.com/technologies/overview/content_language)).
153
  ![miracl](./assets/miracl.jpg)
154
 
155
+ #### LMEB
156
+ On LMEB-Dialogue, a compact embedding model paired with our Nano reranker, which has only 0.27B activated parameters, remains competitive with 7–12B embedding models.
157
+ ![lmeb](./assets/lmeb.jpg)
158
+ ![lmeb_emb](./assets/lmeb_emb.jpg)
159
+
160
  ### Acknowledgements
161
  We sincerely thank `jina-reranker-v3` and `Qwen3-Reranker` for their valuable inspiration and contributions to the reranking community, from which we have learned a lot.
162