Rickey Bowers Jr.

Many instructions can be processed before that one instruction finishes. Did you test it?

This usually happens when a block is not closed. i.e. macro, struc, etc. Try increasing the number of errors output - this usually helps identify the syntax error deeper in...

Reshard the model for the number of GPUs you have: https://gist.github.com/benob/4850a0210b01672175942203aa36d300 Resharding the larger models to one file can improve load times in general.

Perhaps it was intended to be `dist` in reference to `import torch.distributed as dist` at start of file?

Another option is to get the build number from `KUSER_SHARED_DATA.NtBuildNumber`. Not only is it mapped into all processes, but it's the same on 32/64-bit. Reduces to two instructions to exclude...

https://github.com/tgrysztar/fasmg/blob/e7c5fa56dbeb6c22612a9feae3bfd46e0659396f/packages/x86-2/x86-2.inc#L1341-L1351 Also present in the fasm2 refactoring, https://github.com/tgrysztar/fasm2/blob/856414cbec0196d41fda36111cb96284bb628ecb/include/x86-2.inc#L1269-L1286

Weird, I run that test here and I get the correct output. ``` File: lto-directives.dll Format: COFF-x86-64 Arch: x86_64 AddressSize: 64bit Export { Ordinal: 1 Name: entry RVA: 0x1000 }...