[–] 2 points 4 months ago (1 child)

Yup. And if you want to take a small step without major hardware requirements: connect your setup to a paid subscription Mistral or Anthropic API. They allow you to switch off training on your data.

On top of that, the costs are way lower than the normal consumer grade chat subscriptions, and your searches + memory are kept locally (e.g., managed through open webui).

  • source
  • parent
  • context
  • [–] 2 points 4 months ago (3 children)

    "Hey Claude, research for me current research to Nuclear Fusion. What are the biggest hurdles what are the next steps, and how promising is private research" enabling the research feature will give you a report, Fact checked (not clean but ok ish), and all the sources for it.

    Claude will spin up a bunch of workers and search the web, following leads, and so on.

    One of the few actual useful features of AI IMHO

  • source
  • parent
  • context
  • [–] 5 points 4 months ago (2 children)

    I am on Gentoo for it, but everything with a decent rocm should work.

    Have a look for llama-swap, that handles multi head endpoints.

    Also, as you are on a big board, you can quantize yourself, as the BF16 version of qwen has only 72gb.

    I will try and post a full writeup next days. But feel free to dm me, if you need some guidance on quantize or more.

    I am using this fork currently: https://github.com/charlie12345/ROCmFPX

    Stuff happens fast currently, so may be worth to wait a week or two ig you need something super stable, but if you are up for experimenting, that's the way to go

  • source
  • parent
  • context
  • [–] 4 points 4 months ago (5 children)

    For those who want to know more, rough setup:

    • llama-cpp rocmfp4 fork
    • currently custom quantized qwen3.6 35B A3B model, working on publishing
    • be3 embedding and reranker, also GPU
    • gemma4-e4b via FastFlowLM on NPU!
    • OpenWebUI and searxng as docker containers on a Pi currently

    We get 70-100tok/s generation. Four slots with 256k context length each.

    We use a smaller Board with "only" 64GB of shared LPDDR5X. Bottleneck is memory speed, rocmfp4 quants help a lot.

    As soon as I get my imatrix calibration right, I will publish the quantized versions.

    Most existing quantized models are broken. The authors did some not supported stuff (like using a already quantized model and requantize it) that you may get issues with coherence or sudden Chinese words in the output.

    That is not an issue with rocmfp4 but with vibe coders and agent psychosis.

  • source
  • parent
  • context
  • [–] 7 points 4 months ago* (13 children)

    Openwebui+searxng on a AMD strix board.

    Pro: works like a charm, low power consumption, fast, "big" , LLM (running qwen3.6 35B A3B + gemma4 E4B for website summaries and other smaller tasks)

    Con: strix boards start at 2k€, more in USA because of tarrifs

  • source
  • [–] 2 points 4 months ago

    Navidrome user here as well. I use feishen as my Linux and windows desktop client. Big fan of both.

    As others have said, api is just the standard, so it gives you a lot of options

  • source
  • [–] 2 points 4 months ago

    Currently im using Namida which is also great and open-source in contrary to Symfonium.

    Namida supports Local music, Subsonic and Navidrome servers, WebDAV and even YT music if you even need a song that isn't in your collection yet.

    Another alternative (proprietary) is Symfonium

  • source
  • [–] 3 points 4 months ago

    Yep. Dsub is pretty much in feature-complete, maintenance-mode, that is years out of date and hasn't even updated to material 3 UI's which the others have.

    I check the others every few months or so, but they're all still either buggy or missing so many features compared to dsub, that I have to switch back.

  • source
  • parent
  • context
  • [–] 3 points 4 months ago

    Preface: This is what I learnt about Subsonic and such. The informations below may not be fully correct but the core idea is there.

    An API is not a piece of software.

    It is merely a way to define how to make a software "do" an action, like search for a song title or download the song file for streaming/playback.

    Navidrome does not make use of anything from Subsonic, it merely implements (exposes) a Subsonic-compatible API.

    Knowing this information, anyone can run a web request to Navidrome and search for a song, and download/stream it, like they would already do for any other Subsonic-compatible provider:

    You can find a plethora of software that can leverage any Subsonic-compatible API provider to connect to that music library, from dedicated Web UIs to Android and iOS apps, which know how to download and stream songs and play them back.

    This mechanism gives the user the freedom to choose the backend and frontend, and the developers can just implement the part they prefer: for example, the Navidrome team does not have to build an Android or iOS App, whereas App developers don't need to write a serverside application to manage the "real" music archive, decode tags, or even implement a search algorithm, as all this data can be asked via the API.

    Technically, Navidrome wouldn't even need a Web UI.

    Over the years, the Subsonic API has been forked into OpenSubsonic API, which as far as I could tell, merely serves as a way to better define, and to further add new features into, an otherwise closed and possibly reverse-engineered API.

  • source
  • [–] 3 points 4 months ago* (last edited 4 months ago) (1 child)

    Totally agree with you! It's not flashy or modern looking, but everything is where it should be and it all just works exactly as I want it to. I keep others installed on my phone to just to look at if an update improves them, but nothing has yet. Currently installed are ultrasonic, tempo and substream. Non are as good, but all looks better than dsub2000.

  • source
  • parent
  • context
  • [–] 3 points 5 months ago*

    I use Tempus on my phone for our Navidrome server. You can download your music to be available offline, I use it a lot.

    Edit: On their repo it says they have podcast support but I never used it so I don't know how well it works.

  • source
  • view more: ‹ prev next ›