That Kalium with two bonds tho

I like the gemma models bc of the phrasing they use and that they give sources sometimes. The best results though come from llama3 I think. Also openhermes and openchat, which perform well enough for my purposes.

In the beginning i had used microsoft phi, that wasn’t that good though.

Have a look at self hosted alternatives like Ollama in combination with Open-webui. It can be a hassle to set up, or even excruciatingly painful if you never touched a computer before, but it could be worth a try. I use it daily and like it much more than chatgpt to be honest.

Glad i could help ;)

You can get different results, sometimes better sometimes worse, most of the time differently phrased (e.g. the gemma models by google like to do bulletlists and sometimes tell me where they got that information from). There are models specifically trained / finetuned for different tasks (mostly coding, but also writing stories, answering medical questions, telling me what is on a picture, speaking different languages, running on smaller / bigger hardware, etc.). Have a look at ollamas library of models which is outright tiny compared to e.g. huggingface.

Also, i don’t trust OpenAI and others to be confidential with company data or explicit code snippets from work i feed them.

If you’re lucky you just set it to the wrong version, mine uses 10.3.0 (see below).

I tried running the docker container first as well but gave up since there are seperate versions for cuda and rocm which comes packaged with this as well and therefor gets unnecessary big.

I am running it on Fedora natively. I installed it with the setup script from the top of the docs:

curl -fsSL https://ollama.com/install.sh | sh

After that i created a service file (also stated in the linked docs) so that it starts at boot time (so i can just boot my pc and forget it without needing to login).

The crucial part for the GPU in question (RX 6700XT) was this line under the [service] section:

Environment="HSA_OVERRIDE_GFX_VERSION=10.3.0"

As you stated, this sets the environment variable for rocm. Also to be able to reach it from outside of localhost (for my server):

Environment="OLLAMA_HOST=0.0.0.0"

I have my gaming pc running as ollama host when i need it (RX 6700XT with rocm doing the heavy lifting). PC idles at ~50W and draws up to 200W when generating an answer. It is plenty fast though.

My mini pc home server is running openwebui with access to this “ollama instance” but also OpenAIs api when i just need a quick answer and therefor don’t turn on my pc.

All my exes live in texas

It’s a reference to the latest fireship video… I think:

https://youtu.be/AdygBbbEnco

It is great that you could bring up the curiosity to bust the soy myth.

There is no moral consumption of animal products. Many people a lot smarter than both of us (or at least mor dedicated / funded / in their jobs) have made the research and come to this conslusion as well.

The most people who oppose this fact feel attacked at first because it can’t coexist with their own behaviour. It is the same as with every debate where emotion gets brought up as a reasoning though (e.g. refugees, climate change, homeopathy, etc.).

also probably for the planet and a lot of animals.

Better noone tell apple.

Is that Florida in your atlantic ocean or are you just happy to see me?

Yeah i know, i just prefer to use the distros native package manager. That said, i use the jellyfin client from flathub and that one now warns me as well that it depends on qt5.15 (works fine though, since flatpak can have multiple versions of dependencies).

qt was stuck with 5.15 because the telegram app depended on it (sigh). Had to do a dnf upgrade --best --allowerasing for the update to qt6 (and the removal of telegram lol).

But now everything works fine.

I hope it went well :) i was completely ready to go back changing the image tag to v2 but didn’t need to.

Alpine

If you’re ready to tinker a bit i can recommend Ollama for the backend and Open web UI for the frontend. They can also both run on the same machine.

The advantage is that you can use your GPU to compute, which is a lot faster.

I will try cyberpunk one day if its on sale and my pile of shame has gotten smaller.

I made the switch to Heroic from Lutris because the integration is just better. I used both for a while, bc the witcher 3 worked better on the legacy version for me, and heroic didn’t let you choose the (legacy or nextgen), while lutris only had the legacy version. But now you can install any version you want on Heroic (looking at you, every other platform with forced updates). Also, while Lutris downloads the offline installers off of GOG, heroic installs it via the GOG galaxy redistributable. This also makes it possible to sync playtime and savegames, although this is experimental right now. As soon as they start implementing achievements (which i think they have planned) its feature complete for me.

Updates of heroic itself and the games always went fine, although it must be said that the most challenging titles i have on gog right now are witcher 3 and metro exodus.

The codec thing really is a bummer. But thats really one of the few things you would have to do on Fedora while theres plenty of other pitfalls with other distros too. Like an older kernel or having to manually configure drivers for some hardware with Debian, or having to deal with canonicals shenanigans on Ubuntu.

Maybe one of the more niche distros is a better guess for some, like Nobara or Bazzite for gaming.