How to speed it up?

[H9990HH969]

I’m currently deploying it on a 4090D, but the generation speed is even slower than the duration of the generated audio.

here is the logs：
(glmtts_py310) root@141:/home/GLM-TTS-main# python -m tools.gradio_app
/root/miniforge3/envs/glmtts_py310/lib/python3.10/site-packages/jieba/_compat.py:18: UserWarning: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html. The pkg_resources package is slated for removal as early as 2025-11-30. Refrain from using this package or pin to Setuptools<81.
import pkg_resources
Warning: failed to import ttsfrd, use WeTextProcessing instead

• Running on local URL: http://0.0.0.0:8070
• To create a public link, set share=True in launch().
  [load_quantize_encoder] start. model_path='ckpt/speech_tokenizer'
  Configured for 24kHz frontend.
  2025-12-24 22:14:41.816054004 [W:onnxruntime:Default, onnxruntime_pybind_state.cc:640 CreateExecutionProviderInstance] Failed to create CUDAExecutionProvider. Please reference https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html#requirements to ensure all dependencies are met.
  2025-12-24 22:14:42,620 WETEXT INFO building fst for zh_normalizer ...
  INFO: building fst for zh_normalizer ...
  2025-12-24 22:15:45,403 WETEXT INFO done
  INFO: done
  2025-12-24 22:15:45,404 WETEXT INFO fst path: /root/miniforge3/envs/glmtts_py310/lib/python3.10/site-packages/tn/zh_tn_tagger.fst
  INFO: fst path: /root/miniforge3/envs/glmtts_py310/lib/python3.10/site-packages/tn/zh_tn_tagger.fst
  2025-12-24 22:15:45,404 WETEXT INFO /root/miniforge3/envs/glmtts_py310/lib/python3.10/site-packages/tn/zh_tn_verbalizer.fst
  INFO: /root/miniforge3/envs/glmtts_py310/lib/python3.10/site-packages/tn/zh_tn_verbalizer.fst
  2025-12-24 22:15:45,415 WETEXT INFO found existing fst: /root/miniforge3/envs/glmtts_py310/lib/python3.10/site-packages/tn/en_tn_tagger.fst
  INFO: found existing fst: /root/miniforge3/envs/glmtts_py310/lib/python3.10/site-packages/tn/en_tn_tagger.fst
  2025-12-24 22:15:45,415 WETEXT INFO /root/miniforge3/envs/glmtts_py310/lib/python3.10/site-packages/tn/en_tn_verbalizer.fst
  INFO: /root/miniforge3/envs/glmtts_py310/lib/python3.10/site-packages/tn/en_tn_verbalizer.fst
  2025-12-24 22:15:45,415 WETEXT INFO skip building fst for en_normalizer ...
  INFO: skip building fst for en_normalizer ...
  Loading checkpoint shards: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:00<00:00, 22.67it/s]
  Loading HiFT model from ckpt/hift/hift.pt on cuda...