테스트

app.py

@@ -105,11 +105,8 @@ def query_select_summarize_stream(results, query, ai, min_similarity: float = 0.
     else :
         print('일반', results.get('documents'))
         context_parts = [f'검색된 사용자 수는 {len(results.get('ids'))}']
-        docs = [f"{d[d.find('[이름]'): d.find('[', d.find('[이름]')+1)]} {d[d.find('[부서]'): d.find('[', d.find('[부서]')+1)]}" for d in results.get('documents')]
-
-
-
-        context = "\n".join(context_parts + docs)
+        docs = [f"[{d[d.find('[이름]'): d.find('[', d.find('[이름]')+1)]} {d[d.find('[부서]'): d.find('[', d.find('[부서]')+1)]}]" for d in results.get('documents')]
+        context = ",".join(context_parts + docs)
     print(context)
 
     # 모델 로드
@@ -148,7 +145,7 @@ def query_select_summarize_stream(results, query, ai, min_similarity: float = 0.
     generation_kwargs = dict(
         **model_inputs,
         streamer=streamer,
-        max_new_tokens=3000,
+        max_new_tokens=600,
         do_sample=True,
         temperature=0.3,
         top_p=0.9,
@@ -213,7 +210,7 @@ def query_summarize_simple(query: str) :
     # conduct text completion
     generated_ids = model.generate(
         **model_inputs,
-        max_new_tokens=300,
+        max_new_tokens=600,
         do_sample=True,        # ✅ 샘플링 활성화
         temperature=0.3,
         top_p=0.9,