Devcode518
Devcode518
> @shakhat感谢更新!NPU培训整合似乎触及了一些深层次的问题。我将深入研究错误日志,看看是否有潜在的修复或解决方法。会随时通知你的!🛠️ Hello, may I ask if training is now supported on NPU? For example, the Ascend 910A.
> You just need to wait a bit longer for sglang to initialize, sometimes it takes 3-4 minutes to startup, depending on your setup. ProcessLookupError: ERROR:asyncio:Task exception was never retrieved...
> Well, I soved it with removing the DIR /tmp/triton_cache And the TORCH INDUCTOR CACHE DIR like "/tmp/torchinductor_username/", then restart the terminal. And I dont know why, dont konw which...
大佬你好,你这边是拉的它哪个版本的代码?我出现和#93一样的问题,你没有遇到吗?
> 我自己在linux上用3090:我可以在toml中使用torch == 2.5.1,如果您随后将后来的pip安装更改为sglang == 0.4.3并找到https://flashinfer.ai/whl/cu124/torch2.5/flashinfer/而不是../torch2.4/*版本(你需要0.4.3因为0.4.2想要flashinfer 0.1.6和0.2.0是你在给定链接上找到的全部)-所有乱七八糟的结尾的ninja build未能找到遵循通用readme模式的nvcc,但是一个`conda install -c conda-forge cuda-toolkit`似乎平息了这件事。至少足以在本地pdf上运行基本的管道——我没有尝试任何更高级的东西。 Can I ask what your CUDA version is? My computer has CUDA 12.2, and it reports an error...