Christian van der Loo
Christian van der Loo
In the meantime, I suggest trying out ```nx.relabel.convert_node_labels_to_integers```
@s-valent how did you build it? Constructor is not available for M1 yet. EDIT: Never mind. Just installed yarn, ran `yarn && yarn build && yarn dist:mac` to build the...
Well, if there were a macos-arm64 build of Constructor on conda-forge I could install it with Miniforge and build the server component + desktop app fairly easily, and then an...
Sweet!!
`ls -lrt /dev/stdin` outputs `/dev/stdin -> /proc/self/fd/0` Also, can confirm, running in Kubernetes with `stdin: true` and `tty: true`
Thanks for this! Added in `nvim-scrollbar` and was appalled by the drop in performance. Almost uninstalled the plugin if not for this comment.
I don't think this is relevant. Pickled files have long been saved on developer preference; there's no particular reason that they have to use `.pkl` or `.pickle`.
Is there any way to access `var_store` after initializing the default model? Using the zero-shot classification pipeline.
That would help, yes! Any way to access to the underlying var_store without having to rewrite the whole initialization would work.
Hmm... I did make it to a run with MPS, but it seems to be ignoring the `PYTORCH_ENABLE_MPS_FALLBACK=1` environment variable and is failing with TorchError "sparse_grad not supported in MPS...