Lissanro Rayen

> In general when launching SSHFS-Win using net use only the account that launched the file system gets proper access rights. This is what I expected too but not what...

As far as I remember, the issue wan't caused by a space in Windows username, and the UNIX username is without space. Anyway, in newer Windows installation where my Windows...

After reading this https://github.com/winfsp/sshfs-win/issues/347, I have impression there are currently no plans to continue actively develop this project. So I had to move on to something better. In case somebody...

I do not think just using sys.executable is a good idea, this will break the extension for everyone who does not use latest Alpha version. Using sys.executable was the first...

This bug happens when using large "good" file. For example, when I have tried to use 3.9G good file and 3.6G bad file, I got std::bad_alloc error. But when I...

I have also tested locally original GPT-J-6B on my CPU. And the quality of the output is as high as the online version. Unfortunately, quality of the output from the...

I'm using locally original BF16 "slim" weights and they work great so they are sufficiently accurate for inference. But I'm not sure how to use them with your code. After...

> Is it worth me adding this to this PR or making a new one? I suggest creating a separate PR for features unrelated to tensor_split and split_mode, especially given...

> Does anybody know if the tensor_split setting of llama.cpp::server lets you set a zero to exclude a card altogether? Assuming you have three cards (0, 1 and 2) you...

Thanks for quick answer! I will try ggml instead, but does 4-bit ggml have the same quality (perplexity) of output as 4-bit gptq? Or will I need 8-bit ggml to...