13:15:03 <neuroserve> anyone got dalai (https://github.com/cocktailpeanut/dalai) running on smartos or illumos?
13:36:55 <neuroserve> lx zones fail with "request to https://registry.npmjs.org/dalai failed, reason: connect ENETUNREACH 2606:4700::6810:1223:443 - Local (::1:50067)"
13:40:03 <neuroserve> and I don't know how to switch off ipv6 in lx zones...
13:40:25 <nbjoerg> neuroserve: try the gai settings
13:40:34 <nbjoerg> it's not a lx problem at all
13:40:49 <nbjoerg> but f**king idiots not handling multiple DNS records correctly
13:41:59 <nbjoerg> try adding "precedence ::ffff:0:0/96  100
13:42:03 <nbjoerg> " to /etc/gai.conf
14:11:04 <neuroserve> nbjoerg : I'll try that - base-64-trunk zone says btw "Could not find a version that satisfies the requirement torch (from versions: none)" while trying to install torch via pip3
14:20:28 <neuroserve> nbjoerg : that helped - thank you
14:22:12 <nbjoerg> the error itself is not excusable, esp. since that kind of error can happen in a IPv4-only world just as well
14:34:58 <neuroserve> compiling on lx ubuntu 22.04 fails in ggml.c - let's see, whether it compiles in a bhyve instance...
14:41:38 <bahamat> Compilers aren't really using system calls, so if it succeeds in bhyve it's very likely an environmental factor.
14:42:44 <nbjoerg> bahamat: oh boy, let me tell you a story
16:37:30 <neuroserve> bahamat : if it is an environmental issue, I cannot see it - bhyve and lx both use ubuntu 20.04, cc (Ubuntu 9.4.0-1ubuntu1~20.04.1) 9.4.0 and g++ (Ubuntu 9.4.0-1ubuntu1~20.04.1) 9.4.0 - the compilation on lx seems to have problems compiling ggml.c with /usr/lib/gcc/x86_64-linux-gnu/9/include/immintrin.h and /usr/lib/gcc/x86_64-linux-gnu/9/include/f16cintrin.h - from what I can see, those are not different from those on the bhyve instance...
16:53:07 <jperkin> what's the actual error?
16:59:25 <neuroserve> jperkin : here it is in all its glory: https://gist.github.com/neuroserve/bff6df651e9d7eb0bbabcd34572a033a
17:12:40 <jperkin> ok I give up waiting for them to return, someone else can tell them that it'll be related to what gcc thinks is available on the host cpu and the selected options, most likely -mavx
17:12:57 <jperkin> probably something we should look at fixing in what lx is exposing
18:08:22 <neuroserve> jperkin : I came to the same conclusion - the bhyve and the lx vm do not run on the same CN - but I'll check whether the cpus differ. Or is bhyve emulating things here?
18:48:06 <toasterson> A small release for y'all https://crates.io/crates/imgapi very basic but can parse the files and makes a new manifest in preparation to upload but validation of the builder there is still missing