Update README.md
Browse files
README.md
CHANGED
@@ -35,13 +35,13 @@ pipeline_tag: sentence-similarity
|
|
35 |
---
|
36 |
|
37 |
# German Semantic V3
|
38 |
-
|
39 |
|
40 |
The successor ofs [German_Semantic_STS_V2](https://huggingface.co/aari1995/German_Semantic_STS_V2) are here and come with loads of cool new features! While V3 is really knowledge-heavy, V3b is more focused on performance. Feel free to provide feedback on the model and what you would like to see next.
|
41 |
|
42 |
**Note:** To run this model properly, see "Usage".
|
43 |
|
44 |
-
|
45 |
|
46 |
- **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.
|
47 |
- **Sequence length:** Embed up to 8192 tokens (16 times more than V2 and other models)
|
@@ -54,7 +54,7 @@ The successor ofs [German_Semantic_STS_V2](https://huggingface.co/aari1995/Germa
|
|
54 |
|
55 |
(If you are looking for even better performance on tasks, but with a German knowledge-cutoff around 2020, check out [German_Semantic_V3b](https://huggingface.co/aari1995/German_Semantic_V3))
|
56 |
|
57 |
-
|
58 |
|
59 |
This model has some build-in functionality that is rather hidden. To profit from it, use this code:
|
60 |
|
@@ -86,7 +86,7 @@ similarities = model.similarity(embeddings, embeddings)
|
|
86 |
|
87 |
```
|
88 |
|
89 |
-
|
90 |
|
91 |
```
|
92 |
SentenceTransformer(
|
@@ -96,11 +96,11 @@ SentenceTransformer(
|
|
96 |
```
|
97 |
|
98 |
|
99 |
-
|
100 |
|
101 |
Evaluation to come.
|
102 |
|
103 |
-
|
104 |
|
105 |
**Q: Is this Model better than V2?**
|
106 |
|
@@ -126,10 +126,10 @@ Also, V3 uses cls_pooling while V3buses mean_pooling.
|
|
126 |
|
127 |
**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.
|
128 |
|
129 |
-
|
130 |
German_Semantic_V3_Instruct: Guiding your embeddings towards self-selected aspects
|
131 |
|
132 |
-
|
133 |
|
134 |
- To [jinaAI](https://huggingface.co/jinaai) for their BERT implementation that is used, especially ALiBi
|
135 |
- To [deepset](https://huggingface.co/deepset) for the gbert-large, which is a really great model
|
|
|
35 |
---
|
36 |
|
37 |
# German Semantic V3
|
38 |
+
### and [**German_Semantic_V3b**](https://huggingface.co/aari1995/German_Semantic_V3b)
|
39 |
|
40 |
The successor ofs [German_Semantic_STS_V2](https://huggingface.co/aari1995/German_Semantic_STS_V2) are here and come with loads of cool new features! While V3 is really knowledge-heavy, V3b is more focused on performance. Feel free to provide feedback on the model and what you would like to see next.
|
41 |
|
42 |
**Note:** To run this model properly, see "Usage".
|
43 |
|
44 |
+
# Major updates and USPs:
|
45 |
|
46 |
- **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.
|
47 |
- **Sequence length:** Embed up to 8192 tokens (16 times more than V2 and other models)
|
|
|
54 |
|
55 |
(If you are looking for even better performance on tasks, but with a German knowledge-cutoff around 2020, check out [German_Semantic_V3b](https://huggingface.co/aari1995/German_Semantic_V3))
|
56 |
|
57 |
+
# Usage:
|
58 |
|
59 |
This model has some build-in functionality that is rather hidden. To profit from it, use this code:
|
60 |
|
|
|
86 |
|
87 |
```
|
88 |
|
89 |
+
# Full Model Architecture
|
90 |
|
91 |
```
|
92 |
SentenceTransformer(
|
|
|
96 |
```
|
97 |
|
98 |
|
99 |
+
# Evaluation
|
100 |
|
101 |
Evaluation to come.
|
102 |
|
103 |
+
# FAQ
|
104 |
|
105 |
**Q: Is this Model better than V2?**
|
106 |
|
|
|
126 |
|
127 |
**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.
|
128 |
|
129 |
+
# Up next:
|
130 |
German_Semantic_V3_Instruct: Guiding your embeddings towards self-selected aspects
|
131 |
|
132 |
+
# Thank You and Credits
|
133 |
|
134 |
- To [jinaAI](https://huggingface.co/jinaai) for their BERT implementation that is used, especially ALiBi
|
135 |
- To [deepset](https://huggingface.co/deepset) for the gbert-large, which is a really great model
|