aspatic

看到了 load_in_4bit=True所以模型只占了4G 这样是能跑起来 但是你刷榜的时候不会用的是4bit吧? 还是得用16bit 的 FP16吧

另外bithandbyte默认配置的4bit是fp4,英伟达T4这种老卡根本不支持这种量化啊 load_in_4bit (bool, optional, defaults to False) — This flag is used to enable 4-bit quantization by replacing the Linear layers with FP4/NF4 layers from bitsandbytes. 你咋在colab上跑起来的

> I also experienced the same issue when integrating pynini on an arm64 system. I tried modifying the pynini code directly, but it still didn't work properly. I'm wondering if...

但是最好的结果还是应该请官方要么开 cosyvoice-ttsfrd源码 要么给Arm64的预编译安装包 你可以看到他是只给了linux_x86_64的预编译wheel。 cd pretrained_models/CosyVoice-ttsfrd/ ``` unzip resource.zip -d . pip install ttsfrd_dependency-0.1-py3-none-any.whl pip install ttsfrd-0.4.2-cp310-cp310-linux_x86_64.whl ``` 只要官方使能在aarch64上的编译安装cosyvoice-ttsfrd， 就可以完全绕过wetextprocessing 更别提 性能表现也更好 说话停顿更自然。

LongBench v2: https://github.com/THUDM/LongBench/tree/main