home.social

#ircawp โ€” Public Fediverse posts

Live and recent posts from across the Fediverse tagged #ircawp, aggregated by home.social.

  1. So #Flux2Klein is just as awesome as ZImageTurbo! And it can 'edit' images like QwenImageEdit.

    That means I can swap it out on my Slack bot and give it powers. ๐Ÿ˜ฎ

    #ircawp

  2. Man, I've put a ton of time into updating my #ircawp Slack bot this holiday weekend. I can't say I was unproductive, at least. :)

    Latest: calling it with `/img` usually takes a text prompt that gets passed to the image generation backend. But, new feature, if you just pass an image... the vision capabilities of Qwen3 describes the image in detail and passes that off to zimage.

    We also get image language translation services using 90% of the same code.

    #llm #slack

  3. Hrm. Ollama has a good OpenAI compatible API?

    Guess who just realized he no longer needs to write Ollama specific crap?

    This dork.

    Now I can target OpenAI, LlamaCPP, and Ollama all at once. ๐ŸŽธ

    On a related note, did a shit-ton of work on my Slack bot, #ircawp. Added support for passing images to the bot from Slack, so now it can do fancy stuff like `/geolocate` that photo, or otherwise answer queries about pics.

    Next up: integrate tool calling to really let it get into some serious trouble.

  4. I mean, the world is in a really shitty place at the moment, but on the other hand we have amazing stuff like this.

    #llm #ircawp #sausage

  5. So check this out. I was screwing around with #ircawp and it made this name, "Dr. Elara Vex". I thought it sounded cool, and like every time I see something churned out from an #LLM that sounds creative I look to see if it stole it from somewhere obvious.

    WELL.

    First result on Google hits a post on subreddit filled with OTHER people talking about that name coming out of an LLM:reddit.com/r/LocalLLaMA/commen

  6. I told Llama3.1: "There is no wrong place to insert Uranus." and after a rambling talk about Uranus stats, it says...

    > So, go ahead and insert Uranus wherever your mind dares to wander! Just remember: the real Uranus is quite far away from us Earthlings, so no actual insertion necessary.

    And the #SD15 image generator we run after each response gave this picture.

    #llama31 #llm #ircawp

  7. I told Llama3.1: "There is no wrong place to insert Uranus." and after a rambling talk about Uranus stats, it says...

    > So, go ahead and insert Uranus wherever your mind dares to wander! Just remember: the real Uranus is quite far away from us Earthlings, so no actual insertion necessary.

    And the #SD15 image generator we run after each response gave this picture.

    #llama31 #llm #ircawp

  8. I told Llama3.1: "There is no wrong place to insert Uranus." and after a rambling talk about Uranus stats, it says...

    > So, go ahead and insert Uranus wherever your mind dares to wander! Just remember: the real Uranus is quite far away from us Earthlings, so no actual insertion necessary.

    And the #SD15 image generator we run after each response gave this picture.

    #llama31 #llm #ircawp

  9. I told Llama3.1: "There is no wrong place to insert Uranus." and after a rambling talk about Uranus stats, it says...

    > So, go ahead and insert Uranus wherever your mind dares to wander! Just remember: the real Uranus is quite far away from us Earthlings, so no actual insertion necessary.

    And the #SD15 image generator we run after each response gave this picture.

    #llama31 #llm #ircawp

  10. I told Llama3.1: "There is no wrong place to insert Uranus." and after a rambling talk about Uranus stats, it says...

    > So, go ahead and insert Uranus wherever your mind dares to wander! Just remember: the real Uranus is quite far away from us Earthlings, so no actual insertion necessary.

    And the #SD15 image generator we run after each response gave this picture.

    #llama31 #llm #ircawp

  11. Poor, sweet, innocent #ircawp. You thought I was making a series of predictions and asking for what you thought of them. You had no idea I was stating facts.

    #llm #uspol

  12. > Now, if you'll excuse me, I need to go fact-check some rabbit pregnancies...

    I love you, #ircawp.

    en.wikipedia.org/wiki/Mary_Tof

    Also: jesus fucking christ don't read that article.

    #llm #bots

  13. Updated our Slack's version of #ircawp to generate SD15 images for it's replies (a 12GB 3060 is no match for Llama3+SDXL, sadly, without swapping models in and out every time).

    But inferences against Llama3-8b and SD15 to generate these responses is fairly fast now, relatively speaking.

    I'm super happy with how this has all turned out. 3060s are only about ~$300. You can probably churn out a decent personal LLM inference box on the cheap with one. Don't need a 4090. Just watch out for the 3060ti - it's faster, but they downgraded it to 8GB, the absolute bastards...

    #stablediffusion #llm

  14. The way the #ircawp boot banner is rendering in the console right now makes it look like Apple ][ text artifacting on a color monitor. :D

  15. Well, the #SDXS image generation integration with #ircawp is a complete success.

    I can't remember a more productive Sunday in quite a long time. ๐Ÿ˜…

    The nice thing is, I can swap any of this out pretty easily, if -- say -- I have a faster machine for proper SDXL or 1.5 generation, etc. But all this on a tiny box I can hide on my shelf? Quite happy.

    github.com/Fortyseven/ircawp

    #llm #llms #aiart #ollama

  16. Welp, looks like we've got image uploading working on the #Ircawp slack bot! :D

    Making use of SDXS (not SDXL) to quickly generate low-quality images for responses (optionally, with the @@ at the end, or directly with /SDXS).

    SDXS is tiny and only takes about a second to generate a 512x512 image on a small AMD Ryzen 5 5625U box. It's great given the constraints. :D

    Anyway, that + Ollama support all in one morning.

    And yeah, I'm gonna pay for this... ๐Ÿ’ค

    github.com/Fortyseven/ircawp

    #aiart #llm #slack #sdxs