Sentence Similarity
sentence-transformers
ONNX
Safetensors
Chinese
English
modernbert
embeddings
medical
clinical-nlp
healthcare
biomedical
fhir
traditional-chinese
taiwan
rag
text-embeddings-inference
Instructions to use weemed/IlhaEmbed with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- sentence-transformers
How to use weemed/IlhaEmbed with sentence-transformers:
from sentence_transformers import SentenceTransformer model = SentenceTransformer("weemed/IlhaEmbed") sentences = [ "帶狀皰疹", "皮蛇", "高血壓用藥", "114年成健" ] embeddings = model.encode(sentences) similarities = model.similarity(embeddings, embeddings) print(similarities.shape) # [4, 4] - Notebooks
- Google Colab
- Kaggle
docs: add Context-Conditioned usage and MODA Data Provenance to Traditional Chinese section
Browse files
README.md
CHANGED
|
@@ -248,8 +248,24 @@ model = SentenceTransformer("weemed/IlhaEmbed")
|
|
| 248 |
model.encode(["皮蛇", "L-CT", "定期心內門診-戒菸"])
|
| 249 |
```
|
| 250 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 251 |
維度 384,最長 32 個 token。詞表 25.5k(剪過,只留繁體中文和臨床會用到的)。大小 38.5 MB(INT8 ONNX)或 152 MB(fp32 safetensors)。
|
| 252 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 253 |
## 怎麼做的
|
| 254 |
|
| 255 |
基礎模型用 IBM Granite ModernBERT(Apache-2.0),授權寬鬆。用台灣臨床的 teacher 訊號做知識蒸餾,讓學生模型學會把台灣的行話、縮寫、台語,擺到對應的標準概念旁邊。
|
|
|
|
| 248 |
model.encode(["皮蛇", "L-CT", "定期心內門診-戒菸"])
|
| 249 |
```
|
| 250 |
|
| 251 |
+
### 帶 Context 的拉丁縮寫與多義詞動態解歧義
|
| 252 |
+
|
| 253 |
+
純拉丁字母縮寫(如 `MIF`、`CEA`、`eGFR`、`CBC`)以及高度多義詞(如 `鈣化`)本身無漢字特徵且缺乏獨占上下文。透過傳入業務欄位 Context 字串(例如 `f"{column_hint} {fragment}"`),即可在完全不依賴硬編碼字典的情況下精確判讀:
|
| 254 |
+
|
| 255 |
+
```python
|
| 256 |
+
# 帶入欄位上下文 Context 的推論
|
| 257 |
+
query = "hi-checkup 健檢報告 檢體未交 MIF"
|
| 258 |
+
query_vec = model.encode(query, normalize_embeddings=True)
|
| 259 |
+
|
| 260 |
+
# 動態精確對齊至「檢體未交與寄生蟲鏡檢」(得分 ~0.50+,大幅領先勝差 +0.15)
|
| 261 |
+
```
|
| 262 |
+
|
| 263 |
維度 384,最長 32 個 token。詞表 25.5k(剪過,只留繁體中文和臨床會用到的)。大小 38.5 MB(INT8 ONNX)或 152 MB(fp32 safetensors)。
|
| 264 |
|
| 265 |
+
## 資料來源與開放資料標註 (Data Provenance)
|
| 266 |
+
|
| 267 |
+
IlhaEmbed 的蒸餾知識訓練訊號包含數位發展部 (MODA) 開放資料平台之「國教院 13 套學術醫療名詞庫」 (`taic.moda.gov.tw`,共 111,386 筆醫療概念)。原始語料庫遵循政府資料開放授權條款 (MODA Open Data Terms §3.1)。
|
| 268 |
+
|
| 269 |
## 怎麼做的
|
| 270 |
|
| 271 |
基礎模型用 IBM Granite ModernBERT(Apache-2.0),授權寬鬆。用台灣臨床的 teacher 訊號做知識蒸餾,讓學生模型學會把台灣的行話、縮寫、台語,擺到對應的標準概念旁邊。
|