fix: animation replay + chat uses vision model

app.py

@@ -19,7 +19,7 @@ import spaces
 # ── Backend Imports ────────────────────────────────────────────────────
 from backend.app.config import settings
 from backend.app.db.database import db
-from backend.app.core.inference import inference_engine, chat_engine, SCENE_ANALYSIS_PROMPT
+from backend.app.core.inference import inference_engine, SCENE_ANALYSIS_PROMPT
 from backend.app.core.scene_analyzer import SceneAnalyzer
 from backend.app.core.rule_matcher import RuleMatcher
 from backend.app.core.fault_deducer import FaultDeducer
@@ -50,19 +50,6 @@ def gpu_run_inference(image, prompt):
 inference_engine._run_inference = gpu_run_inference
 
 
-# Chat engine GPU wrapper
-_original_chat = chat_engine.chat
-
-
-@spaces.GPU(duration=60)
-def gpu_chat(system_context, user_message):
-    """GPU-accelerated chat inference."""
-    return _original_chat(system_context, user_message)
-
-
-chat_engine.chat = gpu_chat
-
-
 # ── Async helpers ──────────────────────────────────────────────────────
 
 def run_async(coro):
@@ -94,10 +81,6 @@ async def _ensure_init():
         inference_engine.load_model()
     except Exception as e:
         logger.error(f"Vision model load failed: {e}")
-    try:
-        chat_engine.load_model()
-    except Exception as e:
-        logger.error(f"Chat model load failed: {e}")
     _initialized = True
 
 
@@ -619,10 +602,22 @@ SCENE ANALYSES:\n"""
                 if not user_message or not user_message.strip():
                     return history, "", system_ctx
                 ensure_init()
-                if not chat_engine.is_loaded:
-                    chat_engine.load_model()
+                if not inference_engine.is_loaded:
+                    inference_engine.load_model()
                 try:
-                    response = gpu_chat(system_ctx, user_message.strip())
+                    # Use the vision model's text generation capability for chat
+                    chat_prompt = f"""You are TraceScene AI assistant helping with accident analysis.
+
+CONTEXT:
+{system_ctx}
+
+USER QUESTION: {user_message.strip()}
+
+Provide a concise, helpful answer based on the context above."""
+                    # Create a small blank image for the vision model
+                    from PIL import Image as PILImg
+                    blank = PILImg.new('RGB', (64, 64), color=(0, 0, 0))
+                    response = gpu_run_inference(blank, chat_prompt)
                 except Exception as e:
                     response = f"Error: {e}"
                 history = history or []
@@ -697,6 +692,9 @@ SCENE ANALYSES:\n"""
                     num_v = int(num_vehicles)
                 except:
                     pass
+                # Unique ID to force Gradio to re-render on each click (enables replay)
+                import random
+                uid = random.randint(10000, 99999)
                 # Determine animation duration based on severity
                 dur = "3s" if category.lower() == "mild" else "2s" if category.lower() == "medium" else "1.5s"
                 sev_color = "#22c55e" if category.lower() == "mild" else "#f59e0b" if category.lower() == "medium" else "#ef4444"
@@ -737,9 +735,9 @@ SCENE ANALYSES:\n"""
 
                 html = f'''
 <div style="text-align:center; font-family: Inter, Arial, sans-serif;">
-<svg width="700" height="420" viewBox="0 0 700 420" xmlns="http://www.w3.org/2000/svg" style="border:1px solid #444; border-radius:10px; background:#1a1a2e;">
+<svg id="anim_{uid}" width="700" height="420" viewBox="0 0 700 420" xmlns="http://www.w3.org/2000/svg" style="border:1px solid #444; border-radius:10px; background:#1a1a2e;">
   <defs>
-    <radialGradient id="impactGlow" cx="50%" cy="50%" r="50%">
+    <radialGradient id="glow_{uid}" cx="50%" cy="50%" r="50%">
       <stop offset="0%" stop-color="#fbbf24" stop-opacity="0.8"/>
       <stop offset="100%" stop-color="#fbbf24" stop-opacity="0"/>
     </radialGradient>
@@ -757,7 +755,7 @@ SCENE ANALYSES:\n"""
   {v2_svg}
 
   <!-- Impact flash -->
-  <circle cx="340" cy="210" r="0" fill="url(#impactGlow)">
+  <circle cx="340" cy="210" r="0" fill="url(#glow_{uid})">
     <animate attributeName="r" values="0;0;0;0;0;0;0;45;55;0" dur="{dur}" fill="freeze"/>
     <animate attributeName="opacity" values="0;0;0;0;0;0;0;1;0.5;0" dur="{dur}" fill="freeze"/>
   </circle>