Templar 是一款基于 NEAR Protocol 构建的去中心化、非托管型 BTC 借贷平台。它允许用户以 BTC 作为抵押品，无需信任地借入稳定币，旨在释放 BTC 在 NEAR DeFi 生态中的流动性潜力。
查看项目官方Twitter发布的推文
Subnet Showcase: #sn3 Templar @tplr_ai 🚅🚀🔥 The #bittensor $TAO subnet that is doing scalable decentralized training, not theoretically, actually. 0:00 Vision 1:00 Decentralized Training 1:56 Incentive Mechanism 2:42 Miner Stats 3:18 Miner Req 3:33 Roadmap 4:16 How to use https://t.co/mumYfGf00j
子网展示：#sn3 Templar @tplr_ai 🚅🚀🔥 这个#bittensor $TAO子网正在进行可扩展的去中心化训练，不是理论上的，而是实际上的。 0:00 愿景 1:00 去中心化训练 1:56 激励机制 2:42 矿工统计 3:18 矿工要求 3:33 路线图 4:16 如何使用 https://t.co/mumYfGf00j
5️⃣ MYSHELL/TEMPLAR (SN3) - Voice Revolution 🔊 Text-to-speech mastery Decentralized training framework Accessibility focus = huge TAM Early mover advantage in voice AI space
5️⃣ MYSHELL/TEMPLAR (SN3) - 语音革命 🔊 文本到语音的掌握 去中心化训练框架 关注无障碍性 = 巨大的总可寻址市场 语音AI领域的先行者优势
It's 2027 You own your own personal agent — its mind (model weights), memory, and personality are stored in your wallet or local node. That agent can communicate w/ other agents permissionlessly (buy goods/services, do DeFi, co-create projects) via standardized Web3 protocols/rails. If you want, you can fine-tune or retrain your agent locally using (e.g., Bittensor subnets, Gensyn, or FetchAI extensions). If you don't like your AI’s behavior, you can fork the model — not beg a corporation to adjust it. ——————————————— This is a very possible future if Web3 AI established its own moats based on openness, sovereignty, composability, and decentralized infra If players like Nous Research, Pluralis, Templar, Prime Intellect can continuously - Release strong base models, - Allow permissionless fine-tuning (locally or through decentralized networks), - Enable ownership of your own agent personality/memory/data, Then users will prefer these models because they're customizable & ownable, instead of rented from Big AI labs. If decentralized training frameworks/platforms can - Aggregate idle GPUs, - Crowdsource training tasks, - Reduce dependence on AWS/Azure/NVIDIA bottlenecks, We can create a truly decentralized AI economy that no single company can fully dominate And if Web3 builds strong agent standards early (open protocols for agents), agents could naturally prefer Web3 AI networks over centralized API agents (where OpenAI owns your wallet/identity/agent mind). Don't just use AI, own it
这是2027年。你拥有自己的个人代理——它的思维（模型权重）、记忆和个性都存储在你的钱包或本地节点中。该代理可以通过标准化的Web3协议/轨道与其他代理无许可地通信（购买商品/服务、进行DeFi、共同创建项目）。如果你愿意，你可以使用（例如Bittensor子网、Gensyn或FetchAI扩展）在本地微调或重新训练你的代理。如果你不喜欢AI的行为，你可以分叉模型——而不是乞求公司调整它。———————————————如果Web3 AI基于开放性、主权性、可组合性和去中心化基础设施建立了自己的护城河，这是一个非常可能的未来。如果像Nous Research、Pluralis、Templar、Prime Intellect这样的参与者能够持续：- 发布强大的基础模型，- 允许无许可的微调（本地或通过去中心化网络），- 让你拥有自己的代理个性/记忆/数据，那么用户会更喜欢这些模型，因为它们是可定制和可拥有的，而不是从大型AI实验室租用的。如果去中心化的训练框架/平台能够：- 聚合闲置的GPU，- 众包训练任务，- 减少对AWS/Azure/NVIDIA瓶颈的依赖，我们就可以创建一个真正去中心化的AI经济，没有任何一家公司可以完全主导。如果Web3早期建立了强大的代理标准（代理的开放协议），代理可能会自然地更喜欢Web3 AI网络，而不是集中式API代理（OpenAI拥有你的钱包/身份/代理思维）。不要只是使用AI，要拥有它。
Majority of noise in ai/crypto rn is about decentralized training, with the only liquid exposure being Templar on bittensor. They use training algorithms created by the teams mentioned above while being incentivized on mainnet. Nous, prime etc will probably be far more advanced once they go live, but until then, only way to get any exposure afail
当前AI/加密领域的大部分讨论都集中在去中心化训练上，唯一可流通的资产是Bittensor上的Templar。他们使用上述团队创建的训练算法，并在主网上获得激励。Nous、prime等项目上线后可能会先进得多，但在此之前，据我所知这是唯一的参与方式。
One interesting thing happening on @tplr_ai right now: ... aside from the fact that its a permissionless, decentralized and incentivized training run you can contribute to ... Is that its not a deterministic incentive system. Instead, miner are free to liberally optimize their gradients anyway they choose, and are just rewarded on their ability to reduce the loss on the samples, faster and more aggressively than their peers. The added complexity here means the training runs are beginning to move super fast. Faster than we had previously anticipated.
目前@tplr_ai上发生了一件有趣的事情：...除了它是一个无需许可、去中心化和有激励的训练运行，你可以参与贡献...它不是确定性的激励系统。相反，矿工可以自由地以任何他们选择的方式优化他们的梯度，并且仅仅根据他们在样本上减少损失的能力、比同行更快和更积极地进行奖励。增加的复杂性意味着训练运行开始变得非常快。比我们之前预期的要快。
@563defi @brodydotai @Carrot_____1 @TensoraGroup @opentensor @zeussubnet @ReadyAI_ @hippius_subnet @vidaio_ @mogmachine @dotkrueger @404gen_ @BitMindAI @PrimeIntellect @gensynai @NousResearch @PluralisHQ @WeAreTwine Still on H2H ⚔️ @tplr_ai has been getting some love, but so has Distributed Training over on SN38 - each up nicely on the week. These guys allow miners to pre-train a model collaboratively over the internet - the holy grail of decentralized AI, in my opinion https://t.co/GnR6zcJJff
@563defi @brodydotai @Carrot_____1 @TensoraGroup @opentensor @zeussubnet @ReadyAI_ @hippius_subnet @vidaio_ @mogmachine @dotkrueger @404gen_ @BitMindAI @PrimeIntellect @gensynai @NousResearch @PluralisHQ @WeAreTwine 仍在H2H ⚔️中 @tplr_ai 最近获得了一些关注，但SN38上的Distributed Training也是如此 - 本周都有不错的增长。 这些人允许矿工通过互联网协作预训练模型 - 在我看来，这是去中心化AI的圣杯 https://t.co/GnR6zcJJff
emissions to subnets for the past month and change @rayon_labs in the lead with Chutes and Gradients @tplr_ai also catching up with decentralized training https://t.co/KNg7FL16xH
过去一个月子网的排放量和变化 @rayon_labs 以 Chutes and Gradients 领先 @tplr_ai 也通过去中心化训练迎头赶上 https://t.co/KNg7FL16xH
Cool projects here, but none have remotely the amount of real adopt adoption and developer activity as Bittensor… So sadly this piece lacks a lot of credibility. Also, subnet 3 (@tplr_ai) is the only decentralized, incentivized, permissionless training network in the world.
这里有一些很酷的项目，但没有一个能像Bittensor那样拥有真正的采用率和开发者活跃度……因此，这篇文章的可信度大打折扣。此外，子网3（@tplr_ai）是世界上唯一一个去中心化、有激励、无需许可的训练网络。
@JosephJacks_ @tplr_ai Isn't it to be able to sustain such achievement, the subnet miners ought to sell their earning alphas to cover all.costs? And I'm interested in next phase of this decentralized training to be monetized.
@JosephJacks_ @tplr_ai 为了维持这样的成就，子网矿工是否应该出售他们赚取的阿尔法币来覆盖所有成本？而且我对这个去中心化培训的下一阶段如何货币化很感兴趣。
Hundreds of training runs in, the loss is converging and smoothing more and more impressively by the day... this is the worlds only fully permissionless and decentralized training network -- Bittensor subnet 3: @tplr_ai 🔥🔥🔥 https://t.co/zb1FStXpK7
经过数百次训练运行，损失每天都在以令人印象深刻的方式收敛和平滑...这是世界上唯一完全无需许可和去中心化的训练网络——Bittensor subnet 3: @tplr_ai 🔥🔥🔥 https://t.co/zb1FStXpK7