Razi Taj Mazinani
Razi Taj Mazinani
on ubuntu and in a virtual environment (not conda), it was solved by sudo apt install python[version]-dev*. I think the equivalent of it for macos is python-devel. *Header files and...
https://stackoverflow.com/questions/29846087/error-microsoft-visual-c-14-0-is-required-unable-to-find-vcvarsall-bat check out the answers provided here. it's for a different question but can solve the problem of kenlm installation on windows. @TrinhNhatTuyen @TheWorldIsNot @simozhu @chrisma2021 @GNishanth7
download and run visual studio installer and install desktop development with C++, it worked for me. https://visualstudio.microsoft.com/thank-you-downloading-visual-studio/?sku=Community&channel=Release&version=VS2022&source=VSLandingPage&cid=2030&passive=false @quyet12308
hi, thanks for providing open source tools. I'm getting cuda oom error when fine-tuning llama 3 8b on a 12gb gpu with bf16 and qlora config. shouldn't its peak memory...
rtx 3060 and it supports bf16
it happens at the beginning of training. also when tested on llama 3 8b 4bit and mistral 7b awq, faced the same problem.