Hugging Face's logo

aari1995
/
German_Semantic_V3b

Sentence Similarity
sentence-transformers
Safetensors
German
bert
feature-extraction
loss:MatryoshkaLoss
custom_code
text-embeddings-inference
Inference Endpoints
Model card Files Community
3
aari1995 commited on
Commit
e58f964
·
verified ·
1 Parent(s): a15771c

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +8 -8
README.md CHANGED
@@ -31,7 +31,7 @@ The successors of [German_Semantic_STS_V2](https://huggingface.co/aari1995/Germa
31
 
32
  **Note:** To run this model properly, see "Usage".
33
 
34
- ## Major updates and USPs:
35
 
36
  - **Flexibility:** Trained with flexible sequence-length and embedding truncation, flexibility is a core feature of the model. Yet, smaller dimensions bring a minor trade-off in quality.
37
  - **Sequence length:** Embed up to 8192 tokens (16 times more than V2 and other models)
@@ -42,7 +42,7 @@ The successors of [German_Semantic_STS_V2](https://huggingface.co/aari1995/Germa
42
  - **License:** Apache 2.0
43
 
44
 
45
- ## Usage:
46
 
47
  This model has some build-in functionality that is rather hidden. To profit from it, use this code:
48
 
@@ -74,7 +74,7 @@ similarities = model.similarity(embeddings, embeddings)
74
 
75
  ```
76
 
77
- ### Full Model Architecture
78
 
79
  ```
80
  SentenceTransformer(
@@ -84,7 +84,7 @@ SentenceTransformer(
84
  ```
85
 
86
 
87
- ## FAQ
88
 
89
  **Q: Is this Model better than V2?**
90
 
@@ -111,17 +111,17 @@ Another noticable difference is that V3 has a broader cosine_similarity spectrum
111
  **A:** Broadly speaking, when going from 1024 to 512 dimensions, there is very little trade-off (1 percent). When going down to 64 dimensions, you may face a decrease of up to 3 percent.
112
 
113
 
114
- ## Evaluation
115
 
116
  Storage comparison:
117
  ![image/png](https://cdn-uploads.huggingface.co/production/uploads/5f3801ab7e583543386217ac/Aa5WzHanj-DXc86AKxpEz.png)
118
 
119
  Benchmarks: soon.
120
 
121
- ## Up next:
122
- German_Semantic_V3_Instruct: Guiding your embeddings towards self-selected aspects
123
 
124
- ## Thank You and Credits
125
 
126
  - To [jinaAI](https://huggingface.co/jinaai) for their BERT implementation that is used, especially ALiBi
127
  - To [deepset](https://huggingface.co/deepset) for the gbert-large, which is a really great model
 
31
 
32
  **Note:** To run this model properly, see "Usage".
33
 
34
+ # Major updates and USPs:
35
 
36
  - **Flexibility:** Trained with flexible sequence-length and embedding truncation, flexibility is a core feature of the model. Yet, smaller dimensions bring a minor trade-off in quality.
37
  - **Sequence length:** Embed up to 8192 tokens (16 times more than V2 and other models)
 
42
  - **License:** Apache 2.0
43
 
44
 
45
+ # Usage:
46
 
47
  This model has some build-in functionality that is rather hidden. To profit from it, use this code:
48
 
 
74
 
75
  ```
76
 
77
+ ## Full Model Architecture
78
 
79
  ```
80
  SentenceTransformer(
 
84
  ```
85
 
86
 
87
+ # FAQ
88
 
89
  **Q: Is this Model better than V2?**
90
 
 
111
  **A:** Broadly speaking, when going from 1024 to 512 dimensions, there is very little trade-off (1 percent). When going down to 64 dimensions, you may face a decrease of up to 3 percent.
112
 
113
 
114
+ # Evaluation
115
 
116
  Storage comparison:
117
  ![image/png](https://cdn-uploads.huggingface.co/production/uploads/5f3801ab7e583543386217ac/Aa5WzHanj-DXc86AKxpEz.png)
118
 
119
  Benchmarks: soon.
120
 
121
+ # Up next:
122
+ German_Semantic_V3_Instruct: Guiding your embeddings towards self-selected aspects. - planned: 2024.
123
 
124
+ # Thank You and Credits
125
 
126
  - To [jinaAI](https://huggingface.co/jinaai) for their BERT implementation that is used, especially ALiBi
127
  - To [deepset](https://huggingface.co/deepset) for the gbert-large, which is a really great model