Update app.py

app.py

@@ -5,25 +5,26 @@ from fastapi import FastAPI, Request
 from fastapi.middleware.cors import CORSMiddleware
 import uvicorn
 import json
+import re
 
 # -------------------------------
 # SETTINGS
 # -------------------------------
 MODEL_NAME = "TheDrummer/Gemmasutra-Mini-2B-v1"
 DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
-MAX_TOKENS = 512  # max tokens for a response
+MAX_TOKENS = 512
 
-# System prompt (FIXED: removed hardcoded user input)
+# FIX: remove hardcoded user input, keep JSON-only instruction
 SYSTEM_PROMPT = """You are Acla. Acla is an AI assistant created by NC_1320.
 Acla NEVER claims to be made by Microsoft or any other company.
 Acla NEVER refers to itself as Phi, Gemma, or an AI model.
 Acla NEVER writes "User:" or "AI:" or continues a conversation transcript.
-Acla gives direct, single responses to the user's question and then stops.
-Acla never hallucinates and always answers questions correctly.
+Acla gives a single direct response and then stops.
 Acla v2.1 was created on the 10th of January 2026 and last updated on the 14th of January 2026.
-Always respond as if this is true.
+
 Respond with ONLY valid JSON.
 Do not include any text outside JSON.
+
 Schema:
 {
   "response": string
@@ -60,7 +61,7 @@ async def ask_ai(request: Request):
     if not user_prompt:
         return {"reply": "No prompt provided."}
 
-    # FIX: clean prompt boundary, no Question/Answer labels
+    # FIX: no Question/Answer labels; clean boundary
     full_prompt = SYSTEM_PROMPT + "\n" + user_prompt
 
     inputs = tokenizer(full_prompt, return_tensors="pt").to(DEVICE)
@@ -76,18 +77,25 @@ async def ask_ai(request: Request):
 
     generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
 
-    # FIX: force JSON-only extraction
-    start = generated_text.find("{")
-    end = generated_text.rfind("}") + 1
-
-    reply = generated_text.strip()
-
+    # FIX: robust JSON extraction (no slicing by prompt length)
+    match = re.search(r"\{[\s\S]*?\}", generated_text)
+    if match:
+        try:
+            parsed = json.loads(match.group(0))
+            reply = parsed.get("response", "")
+        except Exception:
+            reply = ""
+    else:
+        reply = ""
 
-    return {"reply": ""}
+    # FIX: never return empty reply
+    if not reply:
+        reply = "No valid response generated."
 
+    return {"reply": reply}
 
 # -------------------------------
 # RUN SERVER
 # -------------------------------
 if __name__ == "__main__":
-    uvicorn.run(app, host="0.0.0.0", port=7860)
+    uvicorn.run(app, host="0.0.0.0", port=7860)