Update README.md
Browse files
README.md
CHANGED
@@ -78,6 +78,33 @@ scores = model.predict(
|
|
78 |
print("Scores:", scores)
|
79 |
```
|
80 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
81 |
## HuggingFace transformers
|
82 |
|
83 |
```python
|
|
|
78 |
print("Scores:", scores)
|
79 |
```
|
80 |
|
81 |
+
### SentenceTransformers + onnx の利用
|
82 |
+
|
83 |
+
CPU 環境や arm 環境などで、より高速に動かしたい場合
|
84 |
+
|
85 |
+
```
|
86 |
+
from sentence_transformers import CrossEncoder
|
87 |
+
|
88 |
+
# oxxn のモデルを選ばないと model.onnx が自動で使われる
|
89 |
+
# onnx_filename = None
|
90 |
+
|
91 |
+
# 量子化された最適なモデルを使う場合は、onnx_filename にファイル名を指定する
|
92 |
+
# onnx_filename = "onnx/model_qint8_avx2.onnx"
|
93 |
+
onnx_filename = "onnx/model_qint8_arm64.onnx"
|
94 |
+
|
95 |
+
if onnx_filename:
|
96 |
+
model = CrossEncoder(
|
97 |
+
MODEL_NAME,
|
98 |
+
device="cpu",
|
99 |
+
backend="onnx",
|
100 |
+
model_kwargs={"file_name": onnx_filename},
|
101 |
+
)
|
102 |
+
else:
|
103 |
+
model = CrossEncoder(MODEL_NAME, device="cpu", backend="onnx")
|
104 |
+
|
105 |
+
...
|
106 |
+
```
|
107 |
+
|
108 |
## HuggingFace transformers
|
109 |
|
110 |
```python
|