Out of memory when running CPU version app.py on RK3588

(moss-tts-nano) [user@MOSS-TTS-Nano]$python app.py --device cpu --dtype bfloat16 2026-06-16 20:05:24,567 INFO root:root_path=None 2026-06-16 20:05:24,569 INFO root: loading Nano-TTS checkpoint=OpenMOSS-Team/MOSS-TTS-Nano audio_tokenizer=OpenMOSS-Team/MOSS-Audio-Tokenizer-Nano device=cpu dtype=torch.bfloat16 attn=model_default INFO: Started server process [13185] INFO: Waiting for application startup. INFO: Application startup complete. INFO: Uvicorn running on http://localhost:18083 (Press CTRL+C to quit) 2026-06-16 20:05:25,791 WETEXT INFO building fst for zh_normalizer ... 2026-06-16 20:05:25,791 INFO wetext: building fst for zh_normalizer ... Killed

My RK3588's available memory is less than 6G before running the python script, when buliding fst for zh_normalizer begins, the free memory goes down fast till the script killed by system, why the python script needs so much memory? how to fix it? Is there any parameter about memory? thx

Reproduce the command in app.py on an RK3588 and review the startup logs through zh_normalizer FST construction. Trace where memory rises and check whether the repository exposes a memory-related parameter; done means CPU startup completes within the device's available memory, with a documented workaround or targeted fix.