Jonas Gehring

Hi there! Linux is indeed our preferred platform for development. I agree that the wine setup is not trivial and I'd recommend using either sc-docker (for playing against other bots)...

I assume that you followed https://torchcraft.github.io/TorchCraftAI/docs/install-windows.html and built gflags and glog? Can you double-check whether `../3rdparty/gflags_static.lib` and `../3rdparty/glog/glog.lib` are present?

Yeah, agree re conversion to a `reference_wrapper`. However, if the type in question is copy-constructible it should still be possible to change a property exposed via `prop::as_reference_wrapper` via a variant...

Hi @antonkratz, please take a look at example_completion.py, example_infilling.py, and example_instructions.py for some usage examples. Are there specific parameters you're wondering about?

For including CodeLlama in real applications I would recommend building on top of other open-source inference engines. The repo here serves as a reference implementation, whereas other projects such as...

Hi @fourpartswater, thanks for the notice! Indeed, there was an error in the checksum files for CodeLlama-70b-Python and CodeLlama-70b-Instruct. If you download these models again you should get the updated...

Sorry, we will leave the README as-is.

Hi @andrewgross, indeed `rope_theta` for 70b-Instruct and 70b-Python is 10,000. `max_positioning_embedding` on HuggingFace will be fixed, the correct value is 4096 -- thanks for letting us know!

Good catch @nazarov-yuriy, this is an error in the `checklist.chk` file which we initially provided. This has now been fixed and new downloads will retrieve the corrected checksum file. If...

Can you double-check the contents of the `checksum.chk` in the directory that was downloaded?