Date: Fri, 21 Apr 2023 18:01:45 -0400 From: Aryeh Friedman <aryeh.friedman@gmail.com> To: Mario Marietto <marietto2008@gmail.com> Cc: questions@freebsd.org, FreeBSD Mailing List <freebsd-hackers@freebsd.org> Subject: Re: Installing openAI's GPT-2 Ada AI Language Model Message-ID: <CAGBxaX=mgbx3xjxCyYySR1XGNG013TkebXJ7o8QNak=0HqGyhQ@mail.gmail.com> In-Reply-To: <CA%2B1FSii7q0vZ4v9R94-G-=G%2B7JxYhS_h%2B85ApMce0XJ7MCAc9w@mail.gmail.com> References: <CAGBxaXmhRLk9Lx_ZHeRdoN-K2fRLEhY3cBVtBymmAjd4bBh1OQ@mail.gmail.com> <20230421134120.GA12251@darkbeer.org> <CAGBxaXm%2BSBaiKyvk28%2BKiO1bqP4gfFjMDZz3xB5wbexb6gJUvQ@mail.gmail.com> <CA%2B1FSii7q0vZ4v9R94-G-=G%2B7JxYhS_h%2B85ApMce0XJ7MCAc9w@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, Apr 21, 2023 at 5:37=E2=80=AFPM Mario Marietto <marietto2008@gmail.= com> wrote: > > 1) try to use the linuxulator instead of bhyve (try to remember what Albe= rt Einstein tells about those people that want to solve a problem without c= hanging their perspective : he does not spend nice words for those people) For long term deployment options (eventually I want a VM image people can just download and run) running it on the host in the long term is not a good idea and due to lack of hardware resources (and budget) the only non-production/development platform that I can use is a VM on my production VM host (has 3 other production machines on it). Thus *ANY* solution that does not involve VM's is a non-go. > > 2) actually an nvidia gpu can be passed-thru within linux (and within win= dows 11 if the gpu is amd) See above we don't want to force people to have specific hardware. If I remember right, weren't you the person I was helping late last year to setup pass-thru on bhyve for your nvidia and it turned out to need a bhyve patch (I assume from what you have said this was for the 2 os's you mentioned). If I remember right this was specifically so you could set up GPT-2. > > 3) the tutorial that we are suggesting may work even if you don't want to= use your gpu I was pointed to another tutorial privately that explains how to do it on a VM (I hope ;-)) > > 4) probably you should find the exact version of pytorch that it wants According to chatGPT (and other sources) it will work on anything newer than 1.13 but does not specify anything beyond that.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAGBxaX=mgbx3xjxCyYySR1XGNG013TkebXJ7o8QNak=0HqGyhQ>