Update README.md
Browse files
README.md
CHANGED
|
@@ -28,12 +28,30 @@ This is vocabulary pruned version of [intfloat/multilingual-e5-large](https://hu
|
|
| 28 |
|
| 29 |
Uses only russian and english tokens.
|
| 30 |
|
|
|
|
|
|
|
| 31 |
| | intfloat/multilingual-e5-large | d0rj/e5-large-en-ru |
|
| 32 |
| --- | --- | --- |
|
| 33 |
| Model size (MB) | 2135.82 | 1394.8 |
|
| 34 |
| Params (count) | 559,890,946 | 365,638,14 |
|
| 35 |
| Word embeddings dim | 256,002,048 | 61,749,248 |
|
| 36 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 37 |
## Usage
|
| 38 |
|
| 39 |
### transformers
|
|
|
|
| 28 |
|
| 29 |
Uses only russian and english tokens.
|
| 30 |
|
| 31 |
+
### Size
|
| 32 |
+
|
| 33 |
| | intfloat/multilingual-e5-large | d0rj/e5-large-en-ru |
|
| 34 |
| --- | --- | --- |
|
| 35 |
| Model size (MB) | 2135.82 | 1394.8 |
|
| 36 |
| Params (count) | 559,890,946 | 365,638,14 |
|
| 37 |
| Word embeddings dim | 256,002,048 | 61,749,248 |
|
| 38 |
|
| 39 |
+
### Performance
|
| 40 |
+
|
| 41 |
+
Equal performance on SberQuAD dev benchmark.
|
| 42 |
+
|
| 43 |
+
| Metric on SberQuAD (4122 questions) | intfloat/multilingual-e5-large | d0rj/e5-large-en-ru |
|
| 44 |
+
| --- | --- | --- |
|
| 45 |
+
| recall@3 | 0.787239204269772 | **0.7882096069868996** |
|
| 46 |
+
| map@3 | 0.7230713245997101 | **0.723192624939351** |
|
| 47 |
+
| mrr@3 | 0.7241630276564784 | **0.7243651948892132** |
|
| 48 |
+
| recall@5 | 0.8277535177098496 | **0.8284813197476953** |
|
| 49 |
+
| map@5 | 0.7301603186155587 | **0.7302573588872716** |
|
| 50 |
+
| mrr@5 | 0.7334667637069385 | **0.7335718906679607** |
|
| 51 |
+
| recall@10 | **0.8716642406598738** | 0.871421639980592 |
|
| 52 |
+
| map@10 | **0.7314774917730316** | 0.7313000338687417 |
|
| 53 |
+
| mrr@10 | **0.7392223685527911** | 0.7391814537556898 |
|
| 54 |
+
|
| 55 |
## Usage
|
| 56 |
|
| 57 |
### transformers
|