MaggotHATE

I've been testing p_step for some time now, and it's really interesting. As expected, it gives more variety in answers - but I'd say it also improves creativity and coherence...

Indeed, it seems to be the case of high/low temperature, but I've noticed a significant effect of quantization. I previously tested with q6_k model only (7b mistral finetune), but once...

> Did you already try manually trusting every extension in the browse extensions tab? That's not the point, you're missing **default settings** part. TachiyomiJ2K didn't have custom extensions list in...

> Usually `${Vulkan_GLSLC_EXECUTABLE}` is passed as glslc executable, which should be an executable path, not a command. Did you overwrite that manually? In a portable, i.e. self-contained setup, it is...

> So if you set `--glslc w64devkit\x86_64-w64-mingw32\bin\glslc` does it not find it? Is that because of a relative path issue, or because of std::filesystem? I've already tried it - it...

I think that Acly has correctly pointed out the issue in their [comment](https://github.com/ggml-org/llama.cpp/pull/17144#issuecomment-3516011408): in case of MinGW it should be resolved automatically rather than be treated as an exact executable.

> > I think that Acly has correctly pointed out the issue in their [#17144 (comment)](https://github.com/ggml-org/llama.cpp/pull/17144#issuecomment-3516011408): in case of MinGW it should be resolved automatically rather than be treated as...

> I'll look into resolving the path in cmake and removing the shell from vulkan-shaders-gen. Can you explain what you mean, please? How should I prepare my setup beforehand?

> For the purpose of this issue, we do need logs/info from an unmodified build, otherwise it gets really hard to help. I've just tested compilation with Cmake through the...

> It's probably less painful than maintaining by hand a custom Makefile. Makefile is actually more convenient than Cmake, so that's not the issue here. I can point it to...