Lower model selection smoothing factor from 500 to 200 for faster new model coverage

app.py

@@ -149,7 +149,7 @@ TTS_CACHE_SIZE = int(os.getenv("TTS_CACHE_SIZE", "10"))
 CACHE_AUDIO_SUBDIR = "cache"
 tts_cache = {} # sentence -> {model_a, model_b, audio_a, audio_b, created_at}
 tts_cache_lock = threading.Lock()
-SMOOTHING_FACTOR_MODEL_SELECTION = 500 # For weighted random model selection
+SMOOTHING_FACTOR_MODEL_SELECTION = 200 # For weighted random model selection
 # Increased max_workers to 8 for concurrent generation/refill
 cache_executor = ThreadPoolExecutor(max_workers=8, thread_name_prefix='CacheReplacer')
 all_sentence_texts = [] # Keep the full list available