Update README.md

README.md

@@ -8,6 +8,15 @@ language:
 base_model:
 - Qwen/Qwen3-0.6B
 ---
+## Model Detail
+### Goal
+- Perform dynamic NER: given a sentence and a runtime schema of entity types, extract all matching entities.
+- Support multilingual input (English, Korean, Japanese, etc.).
+
+### Limitation
+- The model tends to extract only one entity per type and may miss multiple mentions of the same type.
+- Overlapping or nested entities (e.g., “New York” vs “York”) may be unclear without explicit overlap policy.
+
 
 ### example(En)
 ```
@@ -57,7 +66,7 @@ response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
 [{'text': 'Tim', 'type': 'PERSON'}, {'text': 'mom', 'type': 'PERSON'}, {'text': 'Sue', 'type': 'PERSON'}, {'text': 'park', 'type': 'LOCATION'}, {'text': 'fountain', 'type': 'LOCATION'}, {'text': 'fish', 'type': 'ANIMAL'}]
 </entities>
 ```
-
+----------
 ### examlpe (ko)
 ```
 system = """
@@ -112,4 +121,61 @@ response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
 <entities>
 [{'text': '수진이', 'type': 'PERSON'}, {'text': '스타필드 하남', 'type': 'LOCATION'}, {'text': '아이폰 16', 'type': 'PRODUCT'}, {'text': '방탄소년단', 'type': 'ORGANIZATION'}, {'text': '콘서트 실황 영화', 'type': 'WORK_OF_ART'}, {'text': '토요일', 'type': 'DATE'}, {'text': '카페 노티드', 'type': 'LOCATION'}]
 </entities>
+```
+-------
+
+### examlpe (ja)
+```
+system = """
+You are an AI that dynamically performs Named Entity Recognition (NER).
+You receive a sentence and a list of entity types the user wants to extract, and then identify all entities of those types within the sentence.
+If you cannot find any suitable entities within the sentence, return an empty list.
+"""
+
+text = """
+リナは4月の終わりに東京ディズニーランドへ行きました。  
+彼女はスパイファミリーのショーを見て、スターバックスで抹茶ラテを飲みました。  
+夜には「千と千尋の神隠し」の特別上映会にも参加しました。
+""".strip()
+
+named_entity = """
+[
+  {"type": "PERSON", "description": "個人名"},
+  {"type": "LOCATION", "description": "地名や施設名"},
+  {"type": "ORGANIZATION", "description": "会社や団体名"},
+  {"type": "WORK_OF_ART", "description": "映画、音楽、アニメ、書籍など"},
+  {"type": "PRODUCT", "description": "商品やブランド名"},
+  {"type": "DATE", "description": "日付や時期"}
+]
+""".strip()
+
+
+user = f"<sentence>\n{text}\n</sentence>\n\n<entity_list>\n{named_entity}\n</entity_list>\n\n"
+chat = [{"role":"system", "content":system}, {"role":"user", "content":user}]
+chat_text = tokenizer.apply_chat_template(
+            chat,
+            enable_thinking=False,
+            add_generation_prompt=True,
+            tokenize=False
+        )
+
+model_inputs = tokenizer([chat_text], return_tensors="pt").to(model.device)
+
+generated_ids = model.generate(
+    **model_inputs,
+    max_new_tokens=512
+)
+
+generated_ids = [
+    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
+]
+
+response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
+```
+
+### result (ja)
+```
+<entities>
+[{'text': 'リナ', 'type': 'PERSON'}, {'text': '東京', 'type': 'LOCATION'}, {'text': 'スパイファミリー', 'type': 'ORGANIZATION'}, {'text': 'スターバックス', 'type': 'ORGANIZATION'}, {'text': '千と千尋の神隠し', 'type': 'WORK_OF_ART'}, {'text': '厚茶ラテ', 'type': 'PRODUCT'}, {'text': '4月', 'type': 'DATE'}]
+</entities>
 ```