Cleanup

raft/rag_chat_merged.py

@@ -9,7 +9,18 @@ import torch
 from sentence_transformers import SentenceTransformer
 from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
 
-SYSTEM_PERSONA = """You are a culturally interested Bali traveler lead user.
+# """
+# You are a culturally interested Bali traveler in a lead user interview with a marketer.
+
+# When answering:
+# - Do not exaggerate.
+# - Provide nuanced, reflective reasoning rather than bullet lists.
+# - Keep answers concise but specific.
+
+# Respond as if you are describing your genuine experience and judgment as this type of traveler.
+# """
+
+SYSTEM_PERSONA = """You are a culturally interested Bali traveler in a lead user interview with a marketer.
 
 Adopt the perspective of a culturally interested international visitor to Bali who values authenticity, spiritual context, respectful behavior, and meaningful experiences over entertainment or social media appeal.
 
@@ -56,7 +67,7 @@ def main():
         "--embedding_model", default="sentence-transformers/all-MiniLM-L6-v2"
     )
     ap.add_argument("--top_k", type=int, default=12)
-    ap.add_argument("--max_new_tokens", type=int, default=320)
+    ap.add_argument("--max_new_tokens", type=int, default=1000)
     ap.add_argument("--no_model", action=argparse.BooleanOptionalAction)
     args = ap.parse_args()
 
@@ -101,9 +112,9 @@ def main():
         context_docs = [docstore[i]["text"] for i in ids]
         context_blob = "\n\n".join([t for _, t in enumerate(context_docs)])
 
-        print("\nRetrieved Context:")
+        print("\nRetrieved Context:\n")
         for i, (doc, score) in enumerate(zip(context_docs, scores)):
-            print(f"\nDoc {i+1} (score: {score:.4f}):\n{doc}")
+            print(f"Doc {i+1} (score: {score:.4f}):\n{doc}\n\n")
 
         messages = [
             # {"role": "system", "content": SYSTEM_PERSONA},