Update README.md
Browse files
README.md
CHANGED
|
@@ -5,15 +5,91 @@ datasets:
|
|
| 5 |
- truro7/vn-law-questions-and-corpus
|
| 6 |
language:
|
| 7 |
- vi
|
| 8 |
-
|
| 9 |
-
- accuracy
|
| 10 |
-
- precision
|
| 11 |
-
- recall
|
| 12 |
base_model: hiieu/halong_embedding
|
| 13 |
-
|
| 14 |
library_name: sentence-transformers
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
tags:
|
| 16 |
- legal
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
|
| 18 |
---
|
| 19 |
|
|
@@ -25,7 +101,3 @@ The model is trained on a dataset of Vietnamese legal questions and correspondin
|
|
| 25 |
|
| 26 |
It uses Matryoshka loss during training and can be truncated to smaller dimensions, allowing for faster comparisons between queries and documents without sacrificing performance.
|
| 27 |
|
| 28 |
-
|
| 29 |
-
---
|
| 30 |
-
license: apache-2.0
|
| 31 |
-
---
|
|
|
|
| 5 |
- truro7/vn-law-questions-and-corpus
|
| 6 |
language:
|
| 7 |
- vi
|
| 8 |
+
|
|
|
|
|
|
|
|
|
|
| 9 |
base_model: hiieu/halong_embedding
|
| 10 |
+
|
| 11 |
library_name: sentence-transformers
|
| 12 |
+
|
| 13 |
+
metrics:
|
| 14 |
+
- cosine_accuracy@1
|
| 15 |
+
- cosine_accuracy@3
|
| 16 |
+
- cosine_accuracy@5
|
| 17 |
+
- cosine_accuracy@10
|
| 18 |
+
- cosine_precision@1
|
| 19 |
+
- cosine_precision@3
|
| 20 |
+
- cosine_precision@5
|
| 21 |
+
- cosine_precision@10
|
| 22 |
+
- cosine_recall@1
|
| 23 |
+
- cosine_recall@3
|
| 24 |
+
- cosine_recall@5
|
| 25 |
+
- cosine_recall@10
|
| 26 |
+
- cosine_ndcg@10
|
| 27 |
+
- cosine_mrr@10
|
| 28 |
+
- cosine_map@100
|
| 29 |
+
|
| 30 |
+
pipeline_tag: sentence-similarity
|
| 31 |
+
|
| 32 |
tags:
|
| 33 |
- legal
|
| 34 |
+
- sentence-transformers
|
| 35 |
+
- sentence-similarity
|
| 36 |
+
- feature-extraction
|
| 37 |
+
- generated_from_trainer
|
| 38 |
+
- loss:MatryoshkaLoss
|
| 39 |
+
- loss:MultipleNegativesRankingLoss
|
| 40 |
+
|
| 41 |
+
model-index:
|
| 42 |
+
- name: Halong Embedding
|
| 43 |
+
results:
|
| 44 |
+
- task:
|
| 45 |
+
type: information-retrieval
|
| 46 |
+
name: Information Retrieval
|
| 47 |
+
metrics:
|
| 48 |
+
- type: cosine_accuracy@1
|
| 49 |
+
value: 0.623
|
| 50 |
+
name: Cosine Accuracy@1
|
| 51 |
+
- type: cosine_accuracy@3
|
| 52 |
+
value: 0.792
|
| 53 |
+
name: Cosine Accuracy@3
|
| 54 |
+
- type: cosine_accuracy@5
|
| 55 |
+
value: 0.851
|
| 56 |
+
name: Cosine Accuracy@5
|
| 57 |
+
- type: cosine_accuracy@10
|
| 58 |
+
value: 0.900
|
| 59 |
+
name: Cosine Accuracy@10
|
| 60 |
+
- type: cosine_precision@1
|
| 61 |
+
value: 0.623
|
| 62 |
+
name: Cosine Precision@1
|
| 63 |
+
- type: cosine_precision@3
|
| 64 |
+
value: 0.412
|
| 65 |
+
name: Cosine Precision@3
|
| 66 |
+
- type: cosine_precision@5
|
| 67 |
+
value: 0.310
|
| 68 |
+
name: Cosine Precision@5
|
| 69 |
+
- type: cosine_precision@10
|
| 70 |
+
value: 0.184
|
| 71 |
+
name: Cosine Precision@10
|
| 72 |
+
- type: cosine_recall@1
|
| 73 |
+
value: 0.353
|
| 74 |
+
name: Cosine Recall@1
|
| 75 |
+
- type: cosine_recall@3
|
| 76 |
+
value: 0.608
|
| 77 |
+
name: Cosine Recall@3
|
| 78 |
+
- type: cosine_recall@5
|
| 79 |
+
value: 0.722
|
| 80 |
+
name: Cosine Recall@5
|
| 81 |
+
- type: cosine_recall@10
|
| 82 |
+
value: 0.823
|
| 83 |
+
name: Cosine Recall@10
|
| 84 |
+
- type: cosine_ndcg@10
|
| 85 |
+
value: 0.706
|
| 86 |
+
name: Cosine Ndcg@10
|
| 87 |
+
- type: cosine_mrr@10
|
| 88 |
+
value: 0.717
|
| 89 |
+
name: Cosine Mrr@10
|
| 90 |
+
- type: cosine_map@100
|
| 91 |
+
value: 0.645
|
| 92 |
+
name: Cosine Map@100
|
| 93 |
|
| 94 |
---
|
| 95 |
|
|
|
|
| 101 |
|
| 102 |
It uses Matryoshka loss during training and can be truncated to smaller dimensions, allowing for faster comparisons between queries and documents without sacrificing performance.
|
| 103 |
|
|
|
|
|
|
|
|
|
|
|
|