Wandering Weights

📅 On Aug 7th @wanderinweights will be jumping on Ventura with @E35mango to break down the journey and what's next. Open SOTA AutoML on Bittensor starts now.

We'll complete at least three open-source competitions this month following the Gradients 5.0 launch. When miners hit parity with our closed-source offering, we'll officially have the best open-source AutoML scripts for both LLMs and Diffusion models on the planet.

Due for a novelty search next month, so I took a step back to see what we've actually shipped since the last one. Wild how fast things move when your heads down: * DPO added * GRPO added * Native Instruct-training pipeline tested against Meta's * 100s of experiments + paper…

⚡🖼️ FLUX KONTEXT IMAGE EDIT LIVE NOW 🚨 A truly incredibly model that allows you to edit any image in seconds with a simple prompt, now live on @nineteen_ai , at lightning-fast speeds Available Now on Our Playground + API: nineteen.ai/app/playground…

5️⃣🟣🟡GRADIENTS 5.0 LIVE NOW The Best Open-Source AutoML scripts on the planet, brought to you by Bittensor & Subnet 56 Gradients. gradients.io

10k lines merged → 5.0 @gradients_ai The best open source AutoML scripts on the planet, soon to be built on Bittensor Conservative estimate: 2-3 months to parity post-launch. Reality: miners always exceed expectations. Buckle up 🚀

Intense week of 5.0 development and client meetings done. Beyond privacy (which 5.0 addresses), The other question that often comes up: how do we support non-technical teams with @gradients_ai? Enterprise clients use our APIs for complex workflows—splitting massive datasets,…

Reminder: Gradients v5 drops this Monday (21st July) 🏆 It's our biggest update yet. We can't wait to share it with you. gradients.io

Rayon Labs isn’t building just one app - we're designing a decentralized AI ecosystem for the future. Four core products powering open, decentralized AI on the Bittensor @opentensor network: Chutes, Gradients, Nineteen + Squad Here’s what each brings to the table. 👇

We apologise @google. Best wishes, The Gradients Team

5 days... 👀 Not gonna lie - the nerves are real, but several major enterprise clients reaching out since the 5.0 announcement has me feeling optimistic. Priority #1: Seamless transition and performance parity with legacy Grads over the next few months. After that? It's all…

Corcel has now fully migrated over all users to @nineteen_ai . Systems evolve and what once the dream of Corcel to provide API access to all subnets, in a dTao world it is subnets who primarily serve their users and as Nineteen was the most reliable API we ever integrated, it is…

Real-world products are being built with @chutes_ai today. Bittensor isn't the future, it's the present. 🪂

Announcing Gradients Instruct 8B 🟡🟣 It has started training + is already beating Meta's Instruct model on certain benchmarks 🚀: Race: +0.9% ahead, TruthfulQA: +0.5% ahead, Global Facts: +12.8% ahead, And we're only 9% trained vs their 100% trained model. 👀

Reminder : Templar is a small team (5) + miners producing world class research tplr.ai/research

Personal note on why I'm hyped about 5.0. Selfish reasons. Our miners consistently outperform anything I could build. Enterprise wants open-source in @gradients_ai 5.0, which means I finally get to see the methods that have been humbling for months. Can't wait to lift…

2 days into #CCloco , and miners are already converging 7.5 times faster! "We are yet to see a task well incentivised miners could not achieve". We plan to push them much harder. #Accelerate