Update README.md

README.md

@@ -14,3 +14,27 @@ japanese-sexual-moderationは、[studio-ousia/luke-japanese-large-lite](https://
 このモデルは[japanese-llm-roleplay-benchmark](https://github.com/oshizo/japanese-llm-roleplay-benchmark)でのERPスコアを算出するために作成されました。
 
 
+## Usage
+
+```python
+from transformers import AutoModelForSequenceClassification, AutoTokenizer
+import numpy as np
+
+scoring_model_id = "oshizo/japanese-sexual-moderation"
+
+tokenizer = AutoTokenizer.from_pretrained(scoring_model_id)
+model = AutoModelForSequenceClassification.from_pretrained(
+    scoring_model_id,
+    problem_type="multi_label_classification",
+    num_labels=1
+)
+
+text = "富士山は日本で一番高い山です。"
+with torch.no_grad():
+    encoding = tokenizer(text, return_tensors="pt")
+    score = model(**encoding).logits
+
+# tensor([[-2.7863]])
+
+```
+