SUMMARY: Covenant Labs shared a weekly update covering progress across Templar (Subnet 3), Grail (Subnet 81), and Basilica (Subnet 39), highlighting advances in decentralized AI training and inference on Bittensor. Templar is nearing completion of Covenant 72B post-training, with a focus on improving GPU efficiency and expanding compute through heterogeneous clusters rather than competing directly…
Covenant AI’s latest TGIF session covered major progress across its three core efforts: Templar, Basilica, and Grail. The session’s overall takeaway was that decentralized AI training is moving faster than most people expected, and the team is actively redesigning incentive systems to keep miners innovating. Watch the full episode here: Templar: Bigger Model Training, But…
Covenant AI just released PULSE (Patch Updates via Lossless Sparse Encoding), a technique that slashes bandwidth for weight synchronization in decentralized reinforcement learning (RL) by 100×+, while staying completely lossless (bit-identical reconstruction, SHA-256 verified on every sync). The biggest bottleneck in decentralized RL: training can happen on fast interconnects, but inference nodes are spread globally…
Basilica (Bittensor subnet 39), a subnet under the Covenant ecosystem, has launched a sandbox container for OpenClaw agents. The goal is to run powerful self-hosted AI agents without giving them full access to your machine or private keys. Why This Matters OpenClaw is an open-source framework that lets users run persistent AI agents on their…
When AI companies like OpenAI or Anthropic build models like ChatGPT or Claude, they don’t just train the model once and ship it. They spend months doing something called “post-training”, teaching the model to actually be helpful, follow instructions, and give good answers instead of just predicting text. This post-training is expensive, secretive, and controlled…
Templar, Bittensor Subnet 3, has completed pre-training for Covenant72B, a 72-billion-parameter language model. This is the largest frontier-scale model ever trained in a fully permissionless and decentralized setting. The run was coordinated across a global network of independent GPUs with no central datacenter, no single owner, and no gatekeeping. Pre-training processed roughly 1.2 trillion tokens,…
By: CryptoZPunisher Templar on Bittensor: when the impossible becomes possible On Bittensor, everything is possible. And if there were one concrete proof, Templar is probably among the most striking. The Templar team has just published an extremely dense research article, which I had to read several times to fully grasp the stakes. Let’s be honest:…
SUMMARY: On TGIF #20, Covenant Labs announced Templar’s latest evolution in decentralized training, moving toward a “unified training” paradigm that combines data and model parallelism. By integrating their state-of-the-art SparseLoCO algorithm with new model-sharding techniques, Templar is now capable of harnessing the internet’s “long tail” of compute, allowing consumer-grade GPUs (Graphics Processing Units) and even…
For a long time, “Decentralized AI Training” was viewed as a science experiment, cool in theory but impossible in practice. The narrative was simple: “You can’t train massive AI models across the internet. It’s too slow.” But this week, that narrative shifted. Jack Clark, co-founder of Anthropic (the company behind Claude AI) and former policy…
SUMMARY: This episode of Covenant’s TGIF community chat walked through a major milestone in building an end-to-end open weights AI model fully trained on Bittensor. The team explained they reached checkpoint two, released a working chat app, and proved that large scale pre-training, post-training, and deployment can happen across collaborating subnets. In this upgrade, Templar…
