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.

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

Im so looking forward to this. When i tried to use tmpfs / ramdisk, the transcoding would simply stop because there was no space left.

Yes, since we have similar gpus you could try the following to run it in a docker container on linux, taken from here and slightly modified:

#!/bin/bash

model=microsoft/phi-2
# share a volume with the Docker container to avoid downloading weights every run
volume=<path-to-your-data-directory>/data

docker run -e HSA_OVERRIDE_GFX_VERSION=10.3.0 -e PYTORCH_ROCM_ARCH="gfx1031" --device /dev/kfd --device /dev/dri --shm-size 1g -p 8080:80 -v $volume:/data ghcr.io/huggingface/text-generation-inference:1.4-rocm --model-id $model

Note how the rocm version has a different tag and that you need to mount your gpu device into the container. The two environment variables are specific to my (any maybe yours also) gpu architecture. It will need a while to download though.

Huggingface TGI is just a piece of software handling the models, like gpt4all. Here is a list of models officially supported by TGI, although they state that you can try different ones as well. You follow the link and look for the files section. The size of the model files (safetensors or pickele binaries) gives a good estimate of how much vram you will need. Sadly this is more than most consumer graphics cards have except for santacoder and microsoft phi.

I tried Huggingface TGI yesterday, but all of the reasonable models need at least 16 gigs of vram. The only model i got working (on a desktop machine with a amd 6700xt gpu) was microsoft phi-2.

Meanwhile 4tb sata ssd is 300€ in germany

It’s true that you shouldn’t open ports to the internet. If you still want your services to be accessible from outside the local network you can install a wireguard server on your thin client that has access to the services you want. And if you really want to harden it you can restrict wireguard clients from ssh and other admin things.

You will need to open one port on the router to your wireguard server though. Wireguard is UDP though and ignores packages without an established connection, so attackers will not even know there is an open port on your router.

Edit: tailscale and zerotier are good external solutions to this as well without needing to open a port at all.

TIL Amazon Echos can act as Zigbee hub. Still not using it lol.

Those examples you listed are not really modern imo. I’m not an UI/UX expert though.

I used Hugo to build my personal website. You can also easily build blogs with it. The difference to the usual approach is that you “code” the website in markdown which makes it super easy. Hugo then generates the html and css for you, which gets statically hosted. Check out the showcases and themes if you’re interested. I used a theme called papermod, it’s pretty common.

I have set the list of people following me and the ones i follow on private now. This seems to be the only useful thing they could scrape. Why would i care if they can see what music i listen to? It’s not like they can see my personal info and link that or is it?

This is what i did. In europe, viable options start at 200€ on ebay (imo). If your use case outgrows one lenovo tiny (which is unlikely since you’re coming from a pi), you can buy more / other tiny pcs / a desktop pc / a server rack and put proxmox on everything for running services inside a cluster.

It sounds like you are accessing it only from your local network, and as long as it stays that way its fine. I have a DNS Server at home, set up a domain name, therefor have certificates from letsencrypt which get automatically renewed for https access, AND STILL I only access it from my local network with no ports open to the www.

If you have your instance openend up to the internet (e.g. by port forwarding from your ISPs router to your nextcloud instance) and have no encryption / no certificate, thats a major flaw and i would advise you to shut it down immediately and close the port.

What you need is a so called File Provider. The linked blog post seems to summarize this pretty good, at least better than the original docs.

You define your path as a “router” and the ip traefik should point to as “service”. Optionally, you can define all kinds of “middlewares”. Example:

# As YAML Configuration File
http:
  routers:
    router1:
      service: myService      <-- Edit: this should be service1, so it uses your defined service below
      middlewares:
        - "foo-add-prefix"
      rule: "Host(`example.com`)"

  middlewares:
    foo-add-prefix:
      addPrefix:
        prefix: "/foo"

  services:
    service1:
      loadBalancer:
        servers:
          - url: "http://127.0.0.1:80"

Im using traefik as reverse proxy for everything, docker container or not.

Bonus: If you own a domain name, you can get https certificates without opening http ports to the outside via dns challenges. You can even get certs for wildcard domains like *.example.com

That’s what I do. I bought a thin client last year, because Raspberry Pi 4s were unavailable to me. So I (fortunately) bought a thin client and got into Proxmox. I have the following containers running on the machine (that i can think of right now):

• Nextcloud (files, notes, contacts, calendars)
• Jellyfin (media streaming)
• Pi hole
• docker host (with home assistant, traefik as reverse proxy for everything, anything else i want lol)
• gitlab instance and kubernetes cluster (implemented in three containers, which is just for learning purposes)

Edit: some of the thing i want to do in the future:

• paperless-ngx (server that a scanner can send files to so your pc doesn’t have to be powered on i believe)
• actual budget (budgeting)
• mediawiki / kiwix (hosting wikipedia dumps)
• some other dashboard than heimdall for links to all my services
• a vpn i can log into from outside of my home network (used to have pivpn on my spare raspberry pi, but want to have it as a container)

Hope i got you some ideas (:

Reminds me of the occasions when iphone customers complained about their battery draining faster / their phones lagging after a software update for years, and just recently apple responded: “you can have battery or features not both lol”.

Regarding features: Usually good software development makes the software more performant over time, not less. But customers are expected to react to excessive DRM measures (like denuvo) or the uprising telemetry hell (like windows 11) with buying more performant hardware. Yields the question what is a (desired) feature and what is a bug, AND what is a cash cow for companies milking their customers.