Mastodon

Germán Martín¿No pensáis que <a href="https://mastodon.social/tags/Gemma3" class="mention hashtag" rel="nofollow noopener" target="_blank">#Gemma3</a> es muy bueno? Que un modelo de tamaño reducido corriendo en local muestre empatía hace dudar de con quién estás hablando

Hacker NewsGemma3 Function Calling<a href="https://ai.google.dev/gemma/docs/capabilities/function-calling" rel="nofollow noopener" translate="no" target="_blank">https://ai.google.dev/gemma/docs/capabilities/function-calling</a><a href="https://mastodon.social/tags/HackerNews" class="mention hashtag" rel="nofollow noopener" target="_blank">#HackerNews</a> <a href="https://mastodon.social/tags/Gemma3" class="mention hashtag" rel="nofollow noopener" target="_blank">#Gemma3</a> <a href="https://mastodon.social/tags/Function" class="mention hashtag" rel="nofollow noopener" target="_blank">#Function</a> <a href="https://mastodon.social/tags/Calling" class="mention hashtag" rel="nofollow noopener" target="_blank">#Calling</a> <a href="https://mastodon.social/tags/AI" class="mention hashtag" rel="nofollow noopener" target="_blank">#AI</a> <a href="https://mastodon.social/tags/Technology" class="mention hashtag" rel="nofollow noopener" target="_blank">#Technology</a> <a href="https://mastodon.social/tags/Functionality" class="mention hashtag" rel="nofollow noopener" target="_blank">#Functionality</a> <a href="https://mastodon.social/tags/Google" class="mention hashtag" rel="nofollow noopener" target="_blank">#Google</a> <a href="https://mastodon.social/tags/Developers" class="mention hashtag" rel="nofollow noopener" target="_blank">#Developers</a>

AmelisI like coding with these AI LLM models! I only have to prompt the AI as if I am a micro managing idiot to get the code that I want. Feels like my previous boss, but he wont manage to spot the provided bugs to update the prompt for a retry.<a href="https://mastodon.green/tags/ai" class="mention hashtag" rel="nofollow noopener" target="_blank">#ai</a> <a href="https://mastodon.green/tags/ollama" class="mention hashtag" rel="nofollow noopener" target="_blank">#ollama</a> <a href="https://mastodon.green/tags/deepseek" class="mention hashtag" rel="nofollow noopener" target="_blank">#deepseek</a> <a href="https://mastodon.green/tags/qwencoder" class="mention hashtag" rel="nofollow noopener" target="_blank">#qwencoder</a> <a href="https://mastodon.green/tags/qodo" class="mention hashtag" rel="nofollow noopener" target="_blank">#qodo</a> <a href="https://mastodon.green/tags/gemma3" class="mention hashtag" rel="nofollow noopener" target="_blank">#gemma3</a>

Hacker NewsFine-tune Google's Gemma 3<a href="https://unsloth.ai/blog/gemma3" rel="nofollow noopener" translate="no" target="_blank">https://unsloth.ai/blog/gemma3</a><a href="https://mastodon.social/tags/HackerNews" class="mention hashtag" rel="nofollow noopener" target="_blank">#HackerNews</a> <a href="https://mastodon.social/tags/Fine" class="mention hashtag" rel="nofollow noopener" target="_blank">#Fine</a>-tune <a href="https://mastodon.social/tags/Gemma" class="mention hashtag" rel="nofollow noopener" target="_blank">#Gemma</a> #3 <a href="https://mastodon.social/tags/Gemma3" class="mention hashtag" rel="nofollow noopener" target="_blank">#Gemma3</a> <a href="https://mastodon.social/tags/AI" class="mention hashtag" rel="nofollow noopener" target="_blank">#AI</a> <a href="https://mastodon.social/tags/MachineLearning" class="mention hashtag" rel="nofollow noopener" target="_blank">#MachineLearning</a> <a href="https://mastodon.social/tags/Google" class="mention hashtag" rel="nofollow noopener" target="_blank">#Google</a> <a href="https://mastodon.social/tags/TechNews" class="mention hashtag" rel="nofollow noopener" target="_blank">#TechNews</a>

michabbb<a href="https://social.vivaldi.net/tags/Mistral" class="mention hashtag" rel="nofollow noopener" target="_blank">#Mistral</a> Small 3.1: SOTA Multimodal <a href="https://social.vivaldi.net/tags/AI" class="mention hashtag" rel="nofollow noopener" target="_blank">#AI</a> with 128k Context Window 🚀<a href="https://social.vivaldi.net/tags/MistralAI" class="mention hashtag" rel="nofollow noopener" target="_blank">#MistralAI</a> releases improved <a href="https://social.vivaldi.net/tags/opensource" class="mention hashtag" rel="nofollow noopener" target="_blank">#opensource</a> model outperforming <a href="https://social.vivaldi.net/tags/Gemma3" class="mention hashtag" rel="nofollow noopener" target="_blank">#Gemma3</a> and <a href="https://social.vivaldi.net/tags/GPT4oMini" class="mention hashtag" rel="nofollow noopener" target="_blank">#GPT4oMini</a> with 150 tokens/sec speed. Features <a href="https://social.vivaldi.net/tags/multimodal" class="mention hashtag" rel="nofollow noopener" target="_blank">#multimodal</a> capabilities under <a href="https://social.vivaldi.net/tags/Apache2" class="mention hashtag" rel="nofollow noopener" target="_blank">#Apache2</a> license.🧵👇<a href="https://social.vivaldi.net/tags/machinelearning" class="mention hashtag" rel="nofollow noopener" target="_blank">#machinelearning</a>

Benjamin Carr, Ph.D. 👨🏻‍💻🧬The <a href="https://hachyderm.io/tags/ollama" class="mention hashtag" rel="nofollow noopener" target="_blank">#ollama</a> <a href="https://hachyderm.io/tags/opensource" class="mention hashtag" rel="nofollow noopener" target="_blank">#opensource</a> <a href="https://hachyderm.io/tags/software" class="mention hashtag" rel="nofollow noopener" target="_blank">#software</a> that makes it easy to run <a href="https://hachyderm.io/tags/Llama3" class="mention hashtag" rel="nofollow noopener" target="_blank">#Llama3</a>, <a href="https://hachyderm.io/tags/DeepSeekR1" class="mention hashtag" rel="nofollow noopener" target="_blank">#DeepSeekR1</a>, <a href="https://hachyderm.io/tags/Gemma3" class="mention hashtag" rel="nofollow noopener" target="_blank">#Gemma3</a>, and other large language models (<a href="https://hachyderm.io/tags/LLM" class="mention hashtag" rel="nofollow noopener" target="_blank">#LLM</a>) is out with its newest release. The ollama software makes it easy to leverage the llama.cpp back-end for running a variety of LLMs and enjoying convenient integration with other desktop software. The new ollama 0.6.2 Release Features Support For <a href="https://hachyderm.io/tags/AMD" class="mention hashtag" rel="nofollow noopener" target="_blank">#AMD</a> <a href="https://hachyderm.io/tags/StrixHalo" class="mention hashtag" rel="nofollow noopener" target="_blank">#StrixHalo</a>, a.k.a. <a href="https://hachyderm.io/tags/RyzenAI" class="mention hashtag" rel="nofollow noopener" target="_blank">#RyzenAI</a> Max+ laptop / SFF desktop SoC. <a href="https://www.phoronix.com/news/ollama-0.6.2" rel="nofollow noopener" translate="no" target="_blank">https://www.phoronix.com/news/ollama-0.6.2</a>

Devin Prater :blind:LLM, local

💓 EV∆ ∆ΠΠ∆ 💓Finally, the setting I've been looking for is disabled and I can move on with life. Oh don't worry, it's not in the usual place of course, every device must be special, if that wasn't ensured by the proclivity in Linux towards "yes there are breaking changes which we might not have documented. submit a bug if you find one, we can't be expected to document releases effectively. what, did you expect that is this a BSD phone? lol,no it's android."Pausing to remind myself that since this is the modern era, finding the answer amongst the vastness of potential data sources and search engine results... the answer was rapidified by asking my local LLM about specific Android version options which are part of the manufacturer's modified UX element tree: > "On a 'TCL NXTPAPER 40 5G' phone, running TCL-UI version 5.0.YGC0, included in the custom settings for "simulated e-ink paper', related to Accessibility Features, where is the floating button for screen color inversion?"and there you have it, one query. I should have defaulted to that by default. also, Gemma3 is very fast on Ollama... now I need to test the quantized version "Fallen-Gemma3-4B-v1g-Q6_K" for a comparison.<a href="https://mastodon.bsd.cafe/tags/Ai" class="mention hashtag" rel="nofollow noopener" target="_blank">#Ai</a> <a href="https://mastodon.bsd.cafe/tags/LLM" class="mention hashtag" rel="nofollow noopener" target="_blank">#LLM</a> <a href="https://mastodon.bsd.cafe/tags/Ollama" class="mention hashtag" rel="nofollow noopener" target="_blank">#Ollama</a> <a href="https://mastodon.bsd.cafe/tags/gemma" class="mention hashtag" rel="nofollow noopener" target="_blank">#gemma</a> <a href="https://mastodon.bsd.cafe/tags/gemma3" class="mention hashtag" rel="nofollow noopener" target="_blank">#gemma3</a> <a href="https://mastodon.bsd.cafe/tags/OSSgemini" class="mention hashtag" rel="nofollow noopener" target="_blank">#OSSgemini</a> <a href="https://mastodon.bsd.cafe/tags/gpu" class="mention hashtag" rel="nofollow noopener" target="_blank">#gpu</a> <a href="https://mastodon.bsd.cafe/tags/android" class="mention hashtag" rel="nofollow noopener" target="_blank">#android</a> <a href="https://mastodon.bsd.cafe/tags/linux" class="mention hashtag" rel="nofollow noopener" target="_blank">#linux</a> <a href="https://mastodon.bsd.cafe/tags/mondaymorning" class="mention hashtag" rel="nofollow noopener" target="_blank">#mondaymorning</a> <a href="https://mastodon.bsd.cafe/tags/nap" class="mention hashtag" rel="nofollow noopener" target="_blank">#nap</a> <a href="https://mastodon.bsd.cafe/tags/naptime" class="mention hashtag" rel="nofollow noopener" target="_blank">#naptime</a> <a href="https://mastodon.bsd.cafe/tags/pleaseNoMoreDebugging" class="mention hashtag" rel="nofollow noopener" target="_blank">#pleaseNoMoreDebugging</a>

ENTER.CO<a href="https://mastodon.social/tags/PorSiTeLoPerdiste" class="mention hashtag" rel="nofollow noopener" target="_blank">#PorSiTeLoPerdiste</a> Google lanza, Gemma 3, la IA más potente que puedes usar en tu propia GPU <a href="https://www.enter.co/especiales/dev/ai/google-lanza-gemma-3-la-ia-mas-potente-que-puedes-usar-en-tu-propia-gpu/?utm_source=dlvr.it&utm_medium=mastodon" rel="nofollow noopener" translate="no" target="_blank">https://www.enter.co/especiales/dev/ai/google-lanza-gemma-3-la-ia-mas-potente-que-puedes-usar-en-tu-propia-gpu/?utm_source=dlvr.it&utm_medium=mastodon</a> <a href="https://mastodon.social/tags/InteligenciaArtificial" class="mention hashtag" rel="nofollow noopener" target="_blank">#InteligenciaArtificial</a> <a href="https://mastodon.social/tags/Gemma3" class="mention hashtag" rel="nofollow noopener" target="_blank">#Gemma3</a> <a href="https://mastodon.social/tags/Google" class="mention hashtag" rel="nofollow noopener" target="_blank">#Google</a>

Karlheinz AgsteinerNot sure if you have noticed it: Google has released Gemma 3, a powerful model that is small enough to run on normal computers.<a href="https://blog.google/technology/developers/gemma-3/" rel="nofollow noopener" translate="no" target="_blank">https://blog.google/technology/developers/gemma-3/</a>I've done some experiments on my Laptop (with a Geforce 3080ti), and am very impressed. I tried to be happy with Llama3, with the Deepseek R1 distills on Llama, with Mistral, but the models that would run on my computer were not in the same league as what you get from ChatGPT or Claude or Deepseek remotely.Gemma changes this for me. So far I let it write 3 smaller pieces of Javascript, analyze a few texts, and it performed slow, but flawlessly. So finally I can move to a "use the local LLM for the 90% default case, and go for the big ones only if the local LLM fails".This way - I use far less CO2 for my LLM tasks - I am in control of my data, nobody can collect my prompts and later sell my profile to ad customers - I am sure the IP of my prompts stay with me - I have the privacy to ask it whatever I want and no server in the US or CN has those data.Interested? If you have a powerful graphiccs card in your PC, it is totally simple:1. install LMStudio from LMStudio.ai 2. in LMStudio, click Discover, and download the Gema3 27b Q4 model 3. ChatIf your graphics card is too small, you might head for the smaller 12b model, but I can't tell you how well it performs.<a href="https://hachyderm.io/tags/LMStudio" class="mention hashtag" rel="nofollow noopener" target="_blank">#LMStudio</a> <a href="https://hachyderm.io/tags/gemma3" class="mention hashtag" rel="nofollow noopener" target="_blank">#gemma3</a> <a href="https://hachyderm.io/tags/gemma" class="mention hashtag" rel="nofollow noopener" target="_blank">#gemma</a> <a href="https://hachyderm.io/tags/chatgpt" class="mention hashtag" rel="nofollow noopener" target="_blank">#chatgpt</a> <a href="https://hachyderm.io/tags/llm" class="mention hashtag" rel="nofollow noopener" target="_blank">#llm</a> <a href="https://hachyderm.io/tags/google" class="mention hashtag" rel="nofollow noopener" target="_blank">#google</a>

Mastokarl 🇺🇦Did a few coding experiments with Gemma 3 local on lmstudio. So far it performs flawless (in terms of capability - on my lowly Geforce 3080ti it is fairly slow, something like 5 tokens per second). But I've got time, and it is mine, running locally, no Billionaire's corporation sees my prompts.For me (privacy nut) this is a big thing, not having to use ChatGPT for everything.<a href="https://mastodon.social/tags/gemma3" class="mention hashtag" rel="nofollow noopener" target="_blank">#gemma3</a> <a href="https://mastodon.social/tags/LMStudio" class="mention hashtag" rel="nofollow noopener" target="_blank">#LMStudio</a> <a href="https://mastodon.social/tags/chatgpt" class="mention hashtag" rel="nofollow noopener" target="_blank">#chatgpt</a>

Lucas Janin 🇨🇦🇫🇷Testing Open WebUi with Gemma:3 on my proxmox mini PC in a LXC. My hardware is limited, 12th Gen Intel Core i5-12450H so I’m only using the 1b (28 token/s) and 4b (11 token/s) version for now.Image description is functioning, but it is slow; it takes 30 seconds to generate this text with the 4b version and 16G allocated for the LXC.Next step, trying this on my Mac M1.<a href="https://mastodon.social/tags/openwebui" class="mention hashtag" rel="nofollow noopener" target="_blank">#openwebui</a> <a href="https://mastodon.social/tags/gemma3" class="mention hashtag" rel="nofollow noopener" target="_blank">#gemma3</a> <a href="https://mastodon.social/tags/selfhosted" class="mention hashtag" rel="nofollow noopener" target="_blank">#selfhosted</a> <a href="https://mastodon.social/tags/selfhost" class="mention hashtag" rel="nofollow noopener" target="_blank">#selfhost</a> <a href="https://mastodon.social/tags/selfhosting" class="mention hashtag" rel="nofollow noopener" target="_blank">#selfhosting</a> <a href="https://mastodon.social/tags/alttext" class="mention hashtag" rel="nofollow noopener" target="_blank">#alttext</a> <a href="https://mastodon.social/tags/ollama" class="mention hashtag" rel="nofollow noopener" target="_blank">#ollama</a> <a href="https://mastodon.social/tags/proxmox" class="mention hashtag" rel="nofollow noopener" target="_blank">#proxmox</a> <a href="https://mastodon.social/tags/lxc" class="mention hashtag" rel="nofollow noopener" target="_blank">#lxc</a> <a href="https://mastodon.social/tags/ia" class="mention hashtag" rel="nofollow noopener" target="_blank">#ia</a>

Mille e Una AvventuraMettiamo alla prova Gemma 3, il nuovo modello linguistico di Google che può essere eseguito su una singola GPU. Il test riguarda il riconoscimento di immagini di videogiochi che il modello linguistico deve poi riprodurre in HTML5. Per non rendere il compito impossibile, scegliamo immagini di videogiochi classici: Pong e Asteroids. Riuscirà il nostro eroe virtuale nell'impresa? <a href="https://mastodon.uno/tags/gemma3" class="mention hashtag" rel="nofollow noopener" target="_blank">#gemma3</a> <a href="https://mastodon.uno/tags/intelligenzaartificiale" class="mention hashtag" rel="nofollow noopener" target="_blank">#intelligenzaartificiale</a> <a href="https://mastodon.uno/tags/programmazione" class="mention hashtag" rel="nofollow noopener" target="_blank">#programmazione</a> <a href="https://mastodon.uno/tags/html5" class="mention hashtag" rel="nofollow noopener" target="_blank">#html5</a> <a href="https://mastodon.uno/tags/javascript" class="mention hashtag" rel="nofollow noopener" target="_blank">#javascript</a> <a href="https://www.youtube.com/watch?v=10B7JyZjGuk" rel="nofollow noopener" translate="no" target="_blank">https://www.youtube.com/watch?v=10B7JyZjGuk</a>

:fckafd: Olli Graf🚟Today I'm trying out <a href="https://nrw.social/tags/gemma3" class="mention hashtag" rel="nofollow noopener" target="_blank">#gemma3</a> on <a href="https://nrw.social/tags/raspberrypi5" class="mention hashtag" rel="nofollow noopener" target="_blank">#raspberrypi5</a> and <a href="https://nrw.social/tags/orangepi5pro" class="mention hashtag" rel="nofollow noopener" target="_blank">#orangepi5pro</a>

Dr. Fortyseven 🥃 █▓▒░Another sort of down side to <a href="https://defcon.social/tags/Gemma3" class="mention hashtag" rel="nofollow noopener" target="_blank">#Gemma3</a>: it kisses your ass SO MUCH. "You were right to question me on that! Good call, sugar tits!"<a href="https://defcon.social/tags/llm" class="mention hashtag" rel="nofollow noopener" target="_blank">#llm</a>

rabimbaLoving how <a href="https://toot.cafe/tags/Gemma3" class="mention hashtag" rel="nofollow noopener" target="_blank">#Gemma3</a> is performant even in a low powered device like this iMac. Bonus it can now understand and describe images too as well as talk in <a href="https://toot.cafe/tags/bengali" class="mention hashtag" rel="nofollow noopener" target="_blank">#bengali</a> <a href="https://bird.makeup/users/googledevexpert" class="u-url mention" rel="nofollow noopener" target="_blank">@googledevexpert</a> <a href="https://bird.makeup/users/googledevs" class="u-url mention" rel="nofollow noopener" target="_blank">@googledevs</a>

Alessio Pomaro🧠 Ho provato <a href="https://mastodon.uno/tags/Gemma3" class="mention hashtag" rel="nofollow noopener" target="_blank">#Gemma3</a> 27B, il nuovo modello open di <a href="https://mastodon.uno/tags/Google" class="mention hashtag" rel="nofollow noopener" target="_blank">#Google</a>, che sembra essere uno dei migliori di questa categoria. ❓ Com'è andata? <a href="https://www.linkedin.com/posts/alessiopomaro_gemma3-google-prompt-activity-7306277042738683904-_K4y" rel="nofollow noopener" translate="no" target="_blank">https://www.linkedin.com/posts/alessiopomaro_gemma3-google-prompt-activity-7306277042738683904-_K4y</a>___ ✉️ 𝗦𝗲 𝘃𝘂𝗼𝗶 𝗿𝗶𝗺𝗮𝗻𝗲𝗿𝗲 𝗮𝗴𝗴𝗶𝗼𝗿𝗻𝗮𝘁𝗼/𝗮 𝘀𝘂 𝗾𝘂𝗲𝘀𝘁𝗲 𝘁𝗲𝗺𝗮𝘁𝗶𝗰𝗵𝗲, 𝗶𝘀𝗰𝗿𝗶𝘃𝗶𝘁𝗶 𝗮𝗹𝗹𝗮 𝗺𝗶𝗮 𝗻𝗲𝘄𝘀𝗹𝗲𝘁𝘁𝗲𝗿: <a href="https://bit.ly/newsletter-alessiopomaro" rel="nofollow noopener" translate="no" target="_blank">https://bit.ly/newsletter-alessiopomaro</a> <a href="https://mastodon.uno/tags/AI" class="mention hashtag" rel="nofollow noopener" target="_blank">#AI</a> <a href="https://mastodon.uno/tags/GenAI" class="mention hashtag" rel="nofollow noopener" target="_blank">#GenAI</a> <a href="https://mastodon.uno/tags/GenerativeAI" class="mention hashtag" rel="nofollow noopener" target="_blank">#GenerativeAI</a> <a href="https://mastodon.uno/tags/IntelligenzaArtificiale" class="mention hashtag" rel="nofollow noopener" target="_blank">#IntelligenzaArtificiale</a> <a href="https://mastodon.uno/tags/LLM" class="mention hashtag" rel="nofollow noopener" target="_blank">#LLM</a>

Dr. Fortyseven 🥃 █▓▒░The <a href="https://defcon.social/tags/Gemma3" class="mention hashtag" rel="nofollow noopener" target="_blank">#Gemma3</a> sure, uh, wants to get me to call the suicide hotline pretty often.The 1B model immediately sensed something off about simply asking "Ass?".<a href="https://defcon.social/tags/llm" class="mention hashtag" rel="nofollow noopener" target="_blank">#llm</a>

Mike StoneSo, I did it. I hooked up the <a href="https://fosstodon.org/tags/HomeAssistant" class="mention hashtag" rel="nofollow noopener" target="_blank">#HomeAssistant</a> Voice to my <a href="https://fosstodon.org/tags/Ollama" class="mention hashtag" rel="nofollow noopener" target="_blank">#Ollama</a> instance. As <a href="https://aus.social/@ianjs" class="u-url mention" rel="nofollow noopener" target="_blank">@ianjs</a> suggested, it's much better at recognizing the intent of my requests. As <a href="https://fosstodon.org/@chris_hayes" class="u-url mention" rel="nofollow noopener" target="_blank">@chris_hayes</a> suggested, I'm using the new <a href="https://fosstodon.org/tags/Gemma3" class="mention hashtag" rel="nofollow noopener" target="_blank">#Gemma3</a> model. It now knows "How's the weather" and "What's the weather" are the same thing, and I get an answer for both. Responses are a little slower than without the LLM, but honestly it's pretty negligible. It's a very little bit slower again if I use local <a href="https://fosstodon.org/tags/Piper" class="mention hashtag" rel="nofollow noopener" target="_blank">#Piper</a> vs HA's cloud service.

Mike StoneTesting out the newly released <a href="https://fosstodon.org/tags/Gemma3" class="mention hashtag" rel="nofollow noopener" target="_blank">#Gemma3</a> model locally on <a href="https://fosstodon.org/tags/ollama" class="mention hashtag" rel="nofollow noopener" target="_blank">#ollama</a>. This is one of the more frustration aspects of these LLMs. It must be said that LLMs are fine for what they are, and what they are is a glorified autocomplete. They have their uses (just like autocomplete does), but if you try to use them outside of their strengths your results are going to be less than reliable.

Recent searches

Search options

Administered by:

Server stats:

#gemma3