Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- en
|
| 4 |
+
metrics:
|
| 5 |
+
- accuracy
|
| 6 |
+
---
|
| 7 |
+
|
| 8 |
+
Fine-Tuned [BLINK](https://github.com/facebookresearch/blink/) CrossEncoder.
|
| 9 |
+
- Base model: https://huggingface.co/UnlikelyAI/crossencoder-wiki-large
|
| 10 |
+
- Training data:
|
| 11 |
+
- 20% (stratified by source dataset) of the following [entity resolution benchmarks](https://console.cloud.google.com/bigquery?ws=!1m4!1m3!3m2!1sunlikelyaiprincipal!2sfixed):
|
| 12 |
+
- handwritten_entity_linking
|
| 13 |
+
- wikibank_entity_linking
|
| 14 |
+
- kilt_entity_linking
|
| 15 |
+
- jobe_entity_linking
|
| 16 |
+
- qald9_entity_linking
|
| 17 |
+
- Training setup:
|
| 18 |
+
- 1 L4 GPU (23GB)
|
| 19 |
+
- batch_size = 1
|
| 20 |
+
- gradient_accumulation_steps = 8
|
| 21 |
+
- type_optimization = "all_encoder_layers" (i.e. ["additional", "bert_model.encoder.layer"])
|
| 22 |
+
- n_epochs = 2
|