Annapurna Labs

The future of AI isn’t just bigger. It’s leaner, faster, and more efficient. On Oct 4, AGI House and AWS are hosting the Small Language Model Build Day — a one-day sprint to prove that smarter design > bigger models. ⚡ Hands-on with AWS Trainium + Hugging Face integrations ⚡ Build production-ready SLM prototypes in just 12 hours ⚡ Explore real-world applications: text-to-SQL, lightweight copilots, reasoning benchmarks Prizes up for grabs: 🥇 1st Place – $3,000 AWS credits, $500 AMZN gift card, and a bonus lunch at the AWS office with the team 🥈 2nd Place – $2,000 AWS credits, $250 AMZN gift card, plus an Alexa 🥉 3rd Place – $1,000 AWS credits + $100 AMZN gift card Speakers + judges include: Ron Diamant (Chief Architect of Trainium @ Amazon Web Services (AWS)) Emily Webber (Principal ML Specialist Solutions Architect @ AWS) Jim Burtoft (Senior Startup Solutions Architect @ AWS) Antoine N. (Head of AI @ Forethought) Linzhi Qi (Co-founder & CTO @ Stealth AI Holding Co) Orian Leitersdorf (Chief Scientist @ Decart AI) Igor Rocha (Co-founder & CPO @ Cigna) 📍 AGI House | 🗓 Oct 4 ⏳ Last chance to apply: Apply Here → https://lnkd.in/gKg7kkMX Rocky Yu Alexa Orent Melissa Daniel Naomi (Yue) Xia Preethika Pavirala Steven Feng

Unlock Telecom AI: NetoAI with Amazon Web Services (AWS) Delivers Telecom Specific LLM with 8x Cost Savings We are proud to announce the successful finetuning and training of our Telecom-Specific Large Language Model (TSLAM 8B) on Amazon SageMaker and AWS Trainium! This achievement marks a significant milestone in bringing highly accurate, domain-specific AI to the telecommunications industry. By leveraging AWS's purpose-built infrastructure, we achieved remarkable results: - Accelerated Training: We completed the finetuning and training process in less than 3 days, enabling rapid iteration and development. - 37% Improved Accuracy: TSLAM 8B reached an impressive 86.2% accuracy, a significant improvement over the base model's performance. - 8x Cheaper: Our training was 8x more cost-effective compared to comparable GPU instances, demonstrating the efficiency and value of AWS Trainium. This success highlights our commitment to delivering cutting-edge, cost-efficient AI solutions tailored for telecom use cases. Ravi Kumar Palepu Vignesh Ethiraj Aravind G #TelecomAI #LLM #Partnership #Innovation #AWS

The Nitro System is one of the main drivers for EC2’s leadership in security and performance. Ever wondered about how it works? I talked about it at re:Invent this week and you can listen to all the innovations here: https://lnkd.in/gbAtPhqh

Building Trainium2 was the most fun I've ever had in my career! It's amazingly fun to build cutting edge technology. It's even more fun to do it with a super-talented and mission-driven team. And it's incredible to see our customers putting everything that we've built to use, and achieving leading performance and price/performance on their workloads and applications. Last week at #reinvent we got a chance to share more about what we've been cooking! :) Hope you like it as much as I do! https://lnkd.in/gwTRpPcJ #trainium2 #deeplearning #aws #anthropic

#postInvent This re:Invent has been such a joy for our #annapurnaalabs team, so much of the work we have been engaged at the past few years finally revealed: 1. Trn2 and Trn2 UltraServers, powered by AWS #Trainium2 chips, are the most powerful instances for deep learning and generative AI on AWS. Enabling up to 30-40% better price performance than current generation GPU-based instances 2. Apple join the main keynote, talking about all the services they run on AWS including their use of #Graviton, #Inferentia and #Trainium2 3. Long list of more Trainium2 launch customers: within the keynote Anthropic, Databricks, Qualcomm, Adobe and poolside. 4. Amazing startups supporting Trainium2: Luma, Brave, NinjaTech, Arcee, 8090, any many more 5. Broad support from popular open source libraries: Hugging Face Optimum, PyTorch Lightening, Ray, Datadog, Scale, Outerbounds, Weights and Biases 6. Google JAX/Trainium partnership: maximizing cross-platform code portability, where customers like #Anthropic using JAX, saying their use of JAX has been valuable for their ability to migrate across platforms, including to AWS Trainium. To learn more, please tune in to our session, where Matthew Johnson (Google DeepMind), Joe Rowell (poolside) and Yang Zhou (Amazon Rufus) joined me to geek out with #Trainium2 https://lnkd.in/gATFR-md

Nitro enables baremetal performance, unparalleled security in the cloud , unlocking speed of innovation at AWS!

From the re:invent keynote stage More than 50% of new CPU capacity landed is #Graviton

#preInvent: make sure you are ready...