Update README.md
Browse files
README.md
CHANGED
|
@@ -7,4 +7,10 @@ Naming pattern:
|
|
| 7 |
|
| 8 |
Actually, models in 1. and 2. are built on top of 3. and 4., respectively.
|
| 9 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
|
|
|
|
| 7 |
|
| 8 |
Actually, models in 1. and 2. are built on top of 3. and 4., respectively.
|
| 9 |
|
| 10 |
+
** NEW **
|
| 11 |
+
|
| 12 |
+
1. `GPL/${dataset}-distilbert-tas-b-gpl-self_miner`: Starting from the [tas-b model](https://huggingface.co/sentence-transformers/msmarco-distilbert-base-tas-b), the models were trained with GPL on the target corpus ${dataset} with the base model itself as the negative miner (here noted as "self_miner"). Query generation setup: Keeping the total number of generated queries = 250K and the default `queries_per_passage` is setup to 3; so if the corpus size is larger than 250K/3=83.3K, the corpus will be truncated to 83.3K; and if the corpus size is below 250K/3=83.3K, the `queries_per_passage` will be increased accordingly to meet the standard (e.g. for 50K-sized corpus, the `queries_per_passage` will be set to 250K/50K=5).
|
| 13 |
+
|
| 14 |
+
|
| 15 |
+
|
| 16 |
|