Hech Interp

Templated from hasktorch-skeleton and haskell-template.

Note, ghc 9.8.4 might not be available in a HLS release yet (you can check this in your ghcup tui). But it is available on HLS master as of Mar 2025. To use this, just

ghcup compile hls --ghc 9.8.4 --git-ref master

and everything should work.

Building

1. Enable the Hasktorch binary cache using Cachix:

   cachix use hasktorch

   This reduces build time significantly on Linux, didn't work on my macbook when I tested it.
2. Launch a Nix shell, which includes GHC with Hasktorch and Haskell Language Server (hls):

   nix develop # You may need to add --extra-experimental-features "nix-command flakes"

3. Try ghci, or cabal build now! Note, stack doesn't seem to work, but that's fine, you don't need to stack ghci because the ghci is in a nix environment with the packages you already need. I.e., you can import Torch in your ghci and it should work.

For VSCode users, in order to get HLS linting support, you can choose to either

1. Launch this repo within the nix shell that was opened by nix develop, or
2. Install direnv, call direnv allow on this repo. When you reopen this repo, direnv will automatically load .envrc.

Apple Specific

Apple uses the MPS backend. There is a toMPS function that deals with this. Try this in your ghci shell:

ghci> Torch.Tensor.toMPS $ Torch.ones' [3,3]

Using IHaskell Notebook

Launch Jupyter Notebook in the nix shell:

ihaskell-notebook