Update app.py

app.py

@@ -1,9 +1,9 @@
-import os, json, re
+import json
 import numpy as np
-from scipy.sparse import hstack
 from sklearn.feature_extraction.text import TfidfVectorizer
 from sklearn.metrics.pairwise import cosine_similarity
 import gradio as gr
+from huggingface_hub import hf_hub_download
 from llama_cpp import Llama
 
 # -------------------- Load Dataset --------------------
@@ -31,23 +31,20 @@ CORPUS = [d["content"]["lo"] for d in DOCS]
 IDS = [d["id"] for d in DOCS]
 ID2DOC = {d["id"]: d for d in DOCS}
 
-word_vec = TfidfVectorizer(ngram_range=(1,2), min_df=1, max_df=0.95, sublinear_tf=True)
-X = word_vec.fit_transform(CORPUS)
+vectorizer = TfidfVectorizer(ngram_range=(1,2), min_df=1, max_df=0.95, sublinear_tf=True)
+X = vectorizer.fit_transform(CORPUS)
 
 # -------------------- Search --------------------
-def search(q, k=3):
-    qv = word_vec.transform([q])
+def search(query, k=3):
+    qv = vectorizer.transform([query])
     sims = cosine_similarity(qv, X)[0]
     idxs = np.argsort(-sims)[:k]
     return [{"id": IDS[i], "score": sims[i]} for i in idxs]
 
-# -------------------- Load Local LLM --------------------
-from huggingface_hub import hf_hub_download
-from llama_cpp import Llama
-
+# -------------------- Load LLM --------------------
 MODEL_PATH = hf_hub_download(
-    repo_id="Qwen/Qwen2.5-3B-Instruct-GGUF", 
-    filename="Qwen2.5-3B-Instruct-Q4_K_M.gguf"  
+    repo_id="Qwen/Qwen2.5-3B-Instruct-GGUF",
+    filename="qwen2.5-3b-instruct-q4_k_m.gguf"   # ✅ ใช้ไฟล์จริง
 )
 
 LLM = Llama(
@@ -57,32 +54,36 @@ LLM = Llama(
     n_gpu_layers=32
 )
 
-
 SYSTEM_RULES = """
 You are a Lao banking assistant for NAYOBY BANK (NBB).
 Answer ONLY from Context. If not found, reply:
 "ຂໍອະໄພ ຂ້ອຍບໍ່ພົບຂໍ້ມູນໃນຖານຄວາມຮູ້."
 """
 
-def build_prompt(q, hits):
+def build_prompt(question, hits):
     ctx = "\n\n".join([f"[{h['id']}] {ID2DOC[h['id']]['content']['lo']}" for h in hits])
-    return f"{SYSTEM_RULES}\n\nContext:\n{ctx}\n\nQuestion:\n{q}\n\nAnswer:"
+    return f"{SYSTEM_RULES}\n\nContext:\n{ctx}\n\nQuestion:\n{question}\n\nAnswer:"
 
-def smart_answer(q):
-    hits = search(q, k=3)
+def smart_answer(message):
+    hits = search(message, k=3)
     if not hits or hits[0]["score"] < 0.1:
         return "ຂໍອະໄພ ບໍ່ພົບຂໍ້ມູນໃນຖານຄວາມຮູ້."
-    prompt = build_prompt(q, hits)
-    out = LLM(prompt, max_tokens=128, temperature=0.2, stop=["Question:"])
-    ans = out["choices"][0]["text"].strip()
-    return ans
+    prompt = build_prompt(message, hits)
+    out = LLM(prompt, max_tokens=128, temperature=0.2)
+    answer = out["choices"][0]["text"].strip()
+    return answer
+
+# -------------------- Gradio Chatbot --------------------
+def respond(message, history):
+    answer = smart_answer(message)
+    history = history + [(message, answer)]
+    return history
 
-# -------------------- Gradio UI --------------------
 with gr.Blocks() as demo:
     gr.Markdown("## 🌾 Lao Chatbot (NBB)")
     chatbot_ui = gr.Chatbot()
     msg = gr.Textbox(placeholder="ພິມຄຳຖາມບ່ອນນີ້...")
-    msg.submit(fn=lambda m, h: (h + [[m, smart_answer(m)]]), inputs=[msg, chatbot_ui], outputs=chatbot_ui)
+    msg.submit(respond, [msg, chatbot_ui], chatbot_ui)
 
 if __name__ == "__main__":
     demo.launch()