GP

dev mockups of what phase 5 will bring into $BREAD game

your baker is yours now. deck them out in a master chef's whites, a royal robe, a dough crown. decorate your bakery, show it off, and visit other bakers' rooms. this is just the start.

Phase 5, version 1 is now live. Get your bakery via customize --> my bakery version 2 and 3 are on the work.

Did you know you can see your overall stats in-game by clicking this button on the UI? you can also review your bases and see what to improve.

1/ Shard protocol for the @c0mputeAI network. github.com/leyten/shard

Time to get @Brimfable running locally. Stay tuned.

Update is in You filled the town, so we made it bigger. Thanks to the latest stability update, the cap is up from 80 to 120 players at once. More of you in one shared world And channels are still coming to lift it even further. cozyville.fun

Behind every feature, every update, every new idea… there’s a lot of code. 😅 HARUKA Farm?

solana:EmcxFTNVDqyLHp11NvwvLZ4D7LKGbG9i7B8RF7dwpump 从根本上改变了人们对消费级硬件和去中心化网络上 AI 推理可能性的认知。 在公共互联网上，多个独立的 4090 之间达到近 25 tok/s，这可不是一个小优化，这是绝无仅有的。 我认为人们还没有意识到 solana:EmcxFTNVDqyLHp11NvwvLZ4D7LKGbG9i7B8RF7dwpump 这一成就的重大意义！！！ solana:EmcxFTNVDqyLHp11NvwvLZ4D7LKGbG9i7B8RF7dwpump 是多GPU在广域网上的推理的规则改变者。 除此之外， solana:EmcxFTNVDqyLHp11NvwvLZ4D7LKGbG9i7B8RF7dwpump 还有完全去中心化和本地AI计算，结合质押机制。这简直是价值数亿美金的项目。 @toly @mert @solana 真的应该看看 @c0mputeAI 在SOL链上构建的东西。

Yep we are earning while being cozy playing a game

Most AI agents today are temporary. You open a chat, ask a question, get a response, and start over tomorrow. We think agents should be more than that. They should have an identity. A personality. A voice. A wallet. A presence that persists beyond a single conversation. We're building toward a future where agents don't just respond. They exist.

Quick heads up on capacity, and what's coming 🌿 For now we've temporarily capped the town at 80 players at once. When it got busier than that, players on phones were crashing, and a town that runs smoothly for everyone comes first. So if you see it full, that's why, and it's only for a few days. The fix is channels, and it's almost here. When the town fills up, a fresh channel of the same Cozyville opens, so you hop straight in with no queue and no waiting line. And here's the important part: it stays ONE shared world. Your account, character, inventory, skills and gold, the market, $COZY and the gold exchange, your friends and your DMs all carry across every channel. The only thing that's per-channel is who you physically see walking around right now. Same map, same town, just a different room. Honest note: at first, you and a friend might land in different channels and not share a screen. Letting you hop straight into a friend's channel is the very next thing we're building. Thanks for bearing with the growing pains while the town grows. More room for everyone, very soon 🐾 cozyville.fun

2/ how shard works: a big model is just a tall stack of layers. shard cuts that stack into slices and hands one slice to each GPU. no single card holds the whole model. a request flows through the slices in order, each one doing its part and passing the result to the next, and the answer comes out the end. so four 4090s that individually top out at small models can pool into one machine that runs a 120B. the hard part is doing this over the open internet instead of inside a datacenter, because every token has to travel the whole chain and that round trip is slow. shard solves it with speculative decoding: a tiny fast model guesses a batch of tokens, the big distributed model checks the whole batch in a single pass, and the verified ones get committed. same network trip, many more tokens per trip. output is identical to running it on one machine, bit for bit.

3/ how it fits c0mpute: right now every c0mpute worker runs a whole model on one GPU, so the network can only serve models that fit on a single card. shard removes that ceiling. workers combine into one virtual big-GPU, and the network can serve frontier-size models that no single contributor could host alone. the more GPUs join, the bigger the models the swarm can run. and it stays true to what c0mpute is. uncensored, models run as-is with no filter in the path. decentralized, anyone joins one GPU with one command and gets assigned a slice. private, no node ever sees the whole model. each GPU in the chain gets paid per token for the work it did. the short version: shard turns a pile of consumer GPUs around the world into a single machine big enough to run frontier models, with no datacenter and no permission needed.

Meet Your Agents

Just got my pengu pet

she's getting her own world.

Over the past few weeks, our development updates have been less frequent than usual. However, we would like to reassure everyone that MPP Layer remains actively developed and continues to move forward. Market conditions and token price fluctuations do not change our long-term vision. Our primary focus has always been building sustainable infrastructure, delivering value to users, and creating tools that developers can rely on. Behind the scenes, we have been continuously working on improvements across the platform, including bug fixes, performance optimizations, and enhancements based on valuable feedback from our users, developers, partners, and community members. Every suggestion, report, and piece of feedback helps us refine the product and shape the future of MPP Layer. We deeply appreciate the support and patience from everyone who continues to believe in what we are building. Development is ongoing, progress is being made, and our commitment remains unchanged: to keep building, improving, and expanding the MPP Layer ecosystem. Thank you for being part of this journey. More updates will be shared as we continue to ship new improvements and features. Build. Improve. Repeat.

A video explaining how the sharding protocol works: (from @onlyzhynx)