qw1319

Results 4 comments of qw1319

nni.common.serializer.PayloadTooLarge: Pickle too large when trying to dump . This might be caused by classes that are not decorated by @nni.trace. Another option is to force bytes pickling and try...

这个问题有解决吗?这边直接运行也看到gpu_offload未提前加载权重 第一步:报错没有activation文件夹; ![image](https://github.com/SJTU-IPADS/PowerInfer/assets/35762142/0fb38d04-9a6f-4e4d-ae17-2b7d9a3c78ab) 这边手动增加activation文件夹(fake)后,执行python依然报错 ![image](https://github.com/SJTU-IPADS/PowerInfer/assets/35762142/b67f9a82-073f-453c-9c2b-5195616fde77)

我遇到和你一样的问题 a100使用系数模型 性能比原始模型还要差很多,测试模型为relullama2-7b

这边同样用a100运行生成的promat貌似是错的 ![image](https://github.com/SJTU-IPADS/PowerInfer/assets/35762142/cc003c3c-1e7c-46ba-a3c5-f4209f33db3b) 同样用cpu跑的结果是正常的 ![image](https://github.com/SJTU-IPADS/PowerInfer/assets/35762142/d940518a-bcec-4ec6-8882-24deb4c908f3) 是否cuda代码不严谨,触发了a100 arch的bug?