Update README.md
Browse files
README.md
CHANGED
@@ -13,6 +13,11 @@ A set of embedding model trained for study embedding quality vs model architectu
|
|
13 |
- **cat-emb-2-256**: 2 layers/H 256/9.7m
|
14 |
- **cat-emb-4-256**: 4 layers/H 256/11.3m
|
15 |
|
|
|
|
|
|
|
|
|
|
|
16 |
### Perf
|
17 |
|
18 |
| MRL dim\Task | BIOSSES | SICK-R | STS12 | STS13 | STS14 | STS15 | STS16 | STSB | SummEval |
|
|
|
13 |
- **cat-emb-2-256**: 2 layers/H 256/9.7m
|
14 |
- **cat-emb-4-256**: 4 layers/H 256/11.3m
|
15 |
|
16 |
+
### Training
|
17 |
+
|
18 |
+
- stage 1: seq 192, batch size 2048, 50k steps, sentence pairs.
|
19 |
+
- stage 2: seq 512, batch size 64, 5k steps, sentence triplets.
|
20 |
+
|
21 |
### Perf
|
22 |
|
23 |
| MRL dim\Task | BIOSSES | SICK-R | STS12 | STS13 | STS14 | STS15 | STS16 | STSB | SummEval |
|