Rename models to wapic-cws{,-base}.wac and move files into model/
Browse files
.gitattributes
CHANGED
|
@@ -36,3 +36,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
| 36 |
base-v5_p0.01.wac filter=lfs diff=lfs merge=lfs -text
|
| 37 |
wapic-20260605.wac filter=lfs diff=lfs merge=lfs -text
|
| 38 |
wapic-20260606-base.wac filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
| 36 |
base-v5_p0.01.wac filter=lfs diff=lfs merge=lfs -text
|
| 37 |
wapic-20260605.wac filter=lfs diff=lfs merge=lfs -text
|
| 38 |
wapic-20260606-base.wac filter=lfs diff=lfs merge=lfs -text
|
| 39 |
+
model/wapic-cws.wac filter=lfs diff=lfs merge=lfs -text
|
| 40 |
+
model/wapic-cws-base.wac filter=lfs diff=lfs merge=lfs -text
|
README.md
CHANGED
|
@@ -21,9 +21,9 @@ tags:
|
|
| 21 |
|
| 22 |
| 文件 | 说明 |
|
| 23 |
|---|---|
|
| 24 |
-
| `wapic-
|
| 25 |
-
| `wapic-
|
| 26 |
-
| `pattern.txt` | 特征模板(unigram + bigram) |
|
| 27 |
|
| 28 |
## 指标
|
| 29 |
|
|
@@ -42,25 +42,25 @@ tags:
|
|
| 42 |
|
| 43 |
```bash
|
| 44 |
# 交互式分词
|
| 45 |
-
./build/wapic -m wapic-
|
| 46 |
# >>> 李镇全今天接受了记者的采访
|
| 47 |
# 李镇全 今天 接受 了 记者 的 采访
|
| 48 |
|
| 49 |
# 批量标注:输入为每行一个字符、空行分句
|
| 50 |
-
./build/wapic test -m wapic-
|
| 51 |
```
|
| 52 |
|
| 53 |
Python 绑定(pybind11,见仓库 `build_py`):
|
| 54 |
|
| 55 |
```python
|
| 56 |
import wapic
|
| 57 |
-
seg = wapic.Segmenter("wapic-
|
| 58 |
seg.cut("中华人民共和国是一个伟大的国家")
|
| 59 |
```
|
| 60 |
|
| 61 |
## 训练
|
| 62 |
|
| 63 |
-
两阶段 warm-start:`wapic-
|
| 64 |
→ 在 name-rich / 现代语料 mix 上 `--init-from` fine-tune → 剪枝。
|
| 65 |
完整配方与数据见数据集仓库 `Ismantic/wapic-cws-data` 的 `RELEASE_TRAINING_DATA.md`。
|
| 66 |
|
|
|
|
| 21 |
|
| 22 |
| 文件 | 说明 |
|
| 23 |
|---|---|
|
| 24 |
+
| `model/wapic-cws.wac` | **主模型**(发布代号 h25_30,48 MB,二进制格式) |
|
| 25 |
+
| `model/wapic-cws-base.wac` | Stage-1 base,warm-start 锚点(仅复现训练时需要) |
|
| 26 |
+
| `model/pattern.txt` | 特征模板(unigram + bigram) |
|
| 27 |
|
| 28 |
## 指标
|
| 29 |
|
|
|
|
| 42 |
|
| 43 |
```bash
|
| 44 |
# 交互式分词
|
| 45 |
+
./build/wapic -m model/wapic-cws.wac
|
| 46 |
# >>> 李镇全今天接受了记者的采访
|
| 47 |
# 李镇全 今天 接受 了 记者 的 采访
|
| 48 |
|
| 49 |
# 批量标注:输入为每行一个字符、空行分句
|
| 50 |
+
./build/wapic test -m model/wapic-cws.wac input_nolabel.txt output_tags.txt
|
| 51 |
```
|
| 52 |
|
| 53 |
Python 绑定(pybind11,见仓库 `build_py`):
|
| 54 |
|
| 55 |
```python
|
| 56 |
import wapic
|
| 57 |
+
seg = wapic.Segmenter("model/wapic-cws.wac")
|
| 58 |
seg.cut("中华人民共和国是一个伟大的国家")
|
| 59 |
```
|
| 60 |
|
| 61 |
## 训练
|
| 62 |
|
| 63 |
+
两阶段 warm-start:`model/wapic-cws-base.wac`(Stage-1,12M LTP 语料)
|
| 64 |
→ 在 name-rich / 现代语料 mix 上 `--init-from` fine-tune → 剪枝。
|
| 65 |
完整配方与数据见数据集仓库 `Ismantic/wapic-cws-data` 的 `RELEASE_TRAINING_DATA.md`。
|
| 66 |
|
pattern.txt → model/pattern.txt
RENAMED
|
File without changes
|
wapic-20260606-base.wac → model/wapic-cws-base.wac
RENAMED
|
File without changes
|
wapic-20260605.wac → model/wapic-cws.wac
RENAMED
|
File without changes
|