Update tts.py

tts.py

@@ -161,7 +161,6 @@ def _init_vllm_engine(model_name, hf_token):
     engine_args = AsyncEngineArgs(
         model=model_name,
         dtype=dtype,
-        quantization="fp8",  # FP8 dynamic quantization — ~1.5x faster on Ada Lovelace (L4)
         gpu_memory_utilization=gpu_mem_util,
         max_model_len=4096,  # Need room for ref audio tokens + generation
         enforce_eager=True,  # Avoid CUDA graph issues on HF Spaces