Today we’re introducing Gemma 4 12B — our latest open model that brings advanced agentic reasoning, vision and audio directly to your laptop.
It delivers performance nearing our larger Gemma models with a much smaller total memory footprint, while being small enough to run locally with just 16GB of VRAM. It’s open and accessible for everyone to use under a permissive Apache 2.0 license.
This is all made possible by our new, unified architecture that removes separate multimodal encoders. Here’s how we did it 🧵
Seven new models launching at Build: let’s go!
Reasoning. Code. Image. Transcribe. Voice.
Built from scratch on a clean data lineage, designed for efficiency, working seamlessly as a family of models
Thread 🧵
#MSBuild
Our First PoC milestone🚀
The team trained and released HimalayaGPT 0.5B-IT, a Nepali-focused smol instruct model built with Karpathy’s nanochat on a GPT-2 style decoder-only architecture.
Trained with Nepali/Devanagari-heavy pretraining, then SFT across instruction, code, math, and tool-use data.
Small model, but a meaningful step toward sovereign AI for Nepal.
Next target: scaling toward 5B models.
Thanks to the entire team and our strategic partner Tarka, and to the NVIDIA DGX Spark stack for powering this training run.
Check out the model Model Releases below↓
I was inspired by this so I wanted to see if Claude Code can get into my Lutron home automation system.
- it found my Lutron controllers on the local wifi network
- checked for open ports, connected, got some metadata and identified the devices and their firmware
- searched the internet, found the pdf for my system
- instructed me on what button to press to pair and get the certificates
- it connected to the system and found all the home devices (lights, shades, HVAC temperature control, motion sensors etc.)
- it turned on and off my kitchen lights to check that things are working (lol!)
I am now vibe coding the home automation master command center, the potential is 🔥.And I'm throwing away the crappy, janky, slow Lutron iOS app I've been using so far. Insanely fun :D :D
#वायुप्रदूषण
जम्मा ३ महिना मात्र हामीले स्वच्छ हावा लिन पाउने रहेछौ, बाँकि ९ महिना प्रदूषित हावामा सास फेर्न बाध्य छौं । किन र कसरी ?...
पुरा भिडियोका लागि तलको लिङ्कमा जानुहोस् ।
https://t.co/Vn5hP36LMr
#NDRRMA#AirPollution#airpollutionawareness#disaster
नेपालको उत्तरपूर्व सीमा नजिकै चीनको डिङ्गयी काउन्टी केन्द्र भएर ७ रेक्टर स्केलको भूकम्प गएको छ।नेपालको ताप्लेजुङ,संखुवासभा,सोलुखुम्बु,दोलखा,सिन्धुपाल्चोक, रसुवा,उत्तरी धादिङमा असर पारेको हुन सक्छ। जिप्रकाहरूबाट सूचना सङ्कलन भैरहेको छ।
तपाईको तिर कस्तो अवस्था छ?जानकारी गराउनुहोला।
An Earthquake of local magnitude (ML ) 7.0 occurred around Dinggye County Of China at 06:50 on 2081/09/23.
NEMRC/DMG
@NDRRMA_Nepal@moha_nepal@NepalOpmcm
We’re thrilled to announce that Bipad Portal, Nepal’s Integrated Disaster Information Management System, has been nominated for the ICT Award 2024 and has proudly reached the semi-finals!
Now, we need your support to help Bipad Portal win the Public Choice ICT Award 2024!
Today is International Day for Disaster Risk Reduction, and the @UN is calling for the empowerment of children and youth for a disaster-free future.
#AreYouReady24#DRRday
A #DigitalFutureForAll is Universal 🔵 Affordable 🔵 Inclusive 🔵 Meaningful 🔵 Sustainable 🔵 Prosperous.
What is your vision of a digital future for all? Share it with us: https://t.co/BYi2l6O55P cc @UNDP@ITU#OurCommonFuture
विपद् सम्बन्धी आधिकारिक सूचना, 'रियल-टाइम अलर्ट', समाचार एवं आवश्यक सुरक्षा निर्देशनहरु उपलब्ध गराउने उद्देश्यले प्राधिकरण र राकुटेन भाइबर (Viber) बीच 'च्यानल' सुरु गर्न समझदारी गरेका छौं । भाइबरमा रहेको 'विपद् सूचना (NDRRMA)' च्यानलमा तपाईंलाई स्वागत छ । आजै Join गर्नुहोस् है ।
मिति २०८१/०४/३२
सोलुखुम्बु जिल्लाको खुम्बु पासाङल्हामु गाउँपालिका ५ थामेको बाढी सम्बन्धी जानकारी
सो क्षेत्रमा हिमपहिरो गएको कारण वा कुनै कारणले थामे खोला जलाधारमा रहेका विभिन्न हिमतालहरू मध्य कुनै हिमताल बिष्फोट (Glacial Lake Outburst Flood) भएको हुन सक्ने देखिन्छ
@NDRRMA_Nepal
Jagged Intelligence
The word I came up with to describe the (strange, unintuitive) fact that state of the art LLMs can both perform extremely impressive tasks (e.g. solve complex math problems) while simultaneously struggle with some very dumb problems.
E.g. example from two days ago - which number is bigger, 9.11 or 9.9? Wrong.
https://t.co/dUrR6wm8GC
or failing to play tic-tac-toe: making non-sensical decisions:
https://t.co/XarwfUBtod
or another common example, failing to count, e.g. the number of times the letter "r" occurs in the word "barrier", ChatGPT-4o claims it's 2:
https://t.co/xpffK2r0pv
The same is true in other modalities. State of the art LLMs can reasonably identify thousands of species of dogs or flowers, but e.g. can't tell if two circles overlap:
https://t.co/HCXxBxosAu
Jagged Intelligence. Some things work extremely well (by human standards) while some things fail catastrophically (again by human standards), and it's not always obvious which is which, though you can develop a bit of intuition over time. Different from humans, where a lot of knowledge and problem solving capabilities are all highly correlated and improve linearly all together, from birth to adulthood.
Personally I think these are not fundamental issues. They demand more work across the stack, including not just scaling. The big one I think is the present lack of "cognitive self-knowledge", which requires more sophisticated approaches in model post-training instead of the naive "imitate human labelers and make it big" solutions that have mostly gotten us this far. For an example of what I'm talking about, see Llama 3.1 paper section on mitigating hallucinations:
https://t.co/pjuxoIOJCY
For now, this is something to be aware of, especially in production settings. Use LLMs for the tasks they are good at but be on a lookout for jagged edges, and keep a human in the loop.
भोली साउन २२ गतेदेखि पर्सि २३ गतेसम्म कोशी, बागमती, गण्डकी, लुम्बिनी र सुदूरपश्चिम प्रदेशका केही तथा कर्णाली र मधेस प्रदेशका थोरै स्थानहरुमा ठूलो पानी पर्ने संभावना भएकाले आवश्यक सतर्कता अपनाउनु हुन अनुरोध छ । साथै अति आवश्यक नपरि घरदेखि टाढाको यात्रा नगर्नु हुन आग्रह गर्दछौं।
Happy #InternationalAIDay! 🎉
It’s a day to remember how #AIoD is revolutionizing the way your AI projects and assets work, with centralized efficiency and customized AI solutions.
Let us know how #AI is making a difference in your world,and let’s celebrate together. 🚀✨
सिमलताल पहिरोमा परी बेपत्ता भएको बस र यात्रुको खोजीका लागि आजदेखि 'Echosounder Device(SONAR)' को प्रयोग गरिएको छ।सशस्त्र प्रहरी बल, नेपालले समन्वय गरी उक्त प्रविधिको प्रयोग मार्फत खोजी कार्य जारी राखेको हो। सोनारले ३ सय मिटर गहिराईसम्मको वस्तुको अवस्थाबारे जानकारी दिन सक्छ।
Exciting News for AI Enthusiasts! The world has started to modernize and advance AI laws. The recent EU AI Law aims to ensure AI systems are safe, transparent, and respectful of fundamental rights. This landmark legislation sets the standard for AI regulation globally, fostering innovation while protecting citizens. Nepal is also making some progress on the AI regulation journey, but more efforts are needed for comprehensive legislation and infrastructure. Authorities are working diligently on this. For more details, check out the EU AI Law here for your reference: https://t.co/u8WrrttGLq
#WeatherForecast shows a possibility of heavy rainfall in #Nepal from tonight. Models are not consistent for the location but are more likely in eastern half of the country.
Clouds in last 12-hr and total rainfall in last 24-hr👇
Next 3-day total rainfall forecast from 2 models👇