Calibrate xlarge reservation from live runtime

app.py

@@ -14,7 +14,6 @@ os.environ.setdefault("PYTORCH_CUDA_ALLOC_CONF", "expandable_segments:True")
 os.environ.setdefault("TOKENIZERS_PARALLELISM", "false")
 
 import glob
-import math
 import subprocess
 import sys
 import time
@@ -246,10 +245,7 @@ def _estimate(
     ):
         return 10
 
-    clip_duration = float(getattr(config, "sound_clip_duration", 30.0))
-    num_clips = max(1, math.ceil(audio_duration / clip_duration))
-    duration = 45 + (num_clips - 1) * 3 + (int(max_new_tokens) / MAX_NEW_TOKENS) * 10
-    return int(min(MAX_GPU_DURATION_SECONDS, math.ceil(duration)))
+    return MAX_GPU_DURATION_SECONDS
 
 
 def _generate(