Update app.py

app.py

@@ -175,7 +175,11 @@ def caption_video(video_path: str) -> str:
         return "No video provided."
 
     T = Timer()
+    
+    t = time.perf_counter()
     model, processor = _load_model_and_processor()
+    if torch.cuda.is_available(): torch.cuda.synchronize()
+    T.mark("load_model_s", time.perf_counter()-t)
 
     print("[ENV] MAX_NEW_TOKENS =", MAX_NEW_TOKENS, flush=True)
     print("[CUDA] available =", torch.cuda.is_available(), flush=True)
@@ -263,6 +267,7 @@ def caption_video(video_path: str) -> str:
         skip_special_tokens=True,
         clean_up_tokenization_spaces=False
     )
+    if torch.cuda.is_available(): torch.cuda.synchronize()
     T.mark("postprocess_s", time.perf_counter()-t)
 
     # print
@@ -280,12 +285,20 @@ def caption_video(video_path: str) -> str:
         except Exception: pass    
 
     # 打印詳細 timing 到日誌（HF Spaces Logs 可見）
-    print({"timings": T.result(), "transcode": tr_info})
+    rt = T.result()
+    known = sum(v for k, v in rt.items() if k != "total_s")
+    other = round(rt["total_s"] - known, 3)
+    print({"timings": rt | {"other_s": other}, "transcode": tr_info}, flush=True)
 
     caption = (output_text[0] if output_text else "").strip()
     if DEBUG_TIMINGS:
-        rt = T.result()
-        caption += f"\n\n[timings] total={rt['total_s']}s, transcode={rt.get('maybe_transcode_s','-')}s, preprocess={rt.get('preprocess_s','-')}s, generate={rt.get('generate_s','-')}s"
+        caption += (
+            f"\n\n[timings] total={rt['total_s']}s, "
+            f"transcode={rt.get('maybe_transcode_s','-')}s, "
+            f"preprocess={rt.get('preprocess_s','-')}s, "
+            f"generate={rt.get('generate_s','-')}s, "
+            f"other={other}s"
+        )
     return caption
 
 # Gradio 介面