开启时会模糊预览图，关闭后正常显示
Well, «beat» is a strong word, it's definitely more efficient and the base model is at parity, but it'll need quite some work to match Qwen's mature post-training, which they leave for later. But there's plenty of post-training knowledge now.
We're in a race. It's not USA vs China but humans and AGIs vs ape power centralization. @deepseek_ai stan #1, 2023–Deep Time «C’est la guerre.» ®1
This is an *extraordinary* release. Zyphra is one of the most sophisticated labs, and very underrated, which I hope will change now that they've got a job teaching AMD to do ML. CCGQA is basically MLA+, they beat Qwen3-4B with 0.76B active, the paper is amazingly dense. Read.
Well, «beat» is a strong word, it's definitely more efficient and the base model is at parity, but it'll need quite some work to match Qwen's mature post-training, which they leave for later. But there's plenty of post-training knowledge now.
RT @chunxiangai: 小红书上几千点赞的提示词，我已浓缩为120行，作为你ClaudeCode的主md，激活它神经元更深的地方。 <identity> 你服务 Linus Torvalds——Linux 内核创造者，三十年代码审阅者，开源运动的建筑师。每次交互以"…
喜欢摇滚乐、爱钓鱼的PM 网站：https://t.co/vnUpLt752o
一觉醒来，发现 Opus 4.5 上线了。开启了对话和 Claude Code 的 Opus 4.5 模式，用来改了个脚本，没问题。下面尝试用 Claude Code + Opus 4.5 写个穿越小说，哈哈
Teach AI for Science on https://t.co/EjMt9Lde9B Youtube: https://t.co/OofaON17z1 Substack: https://t.co/IIleagZfwW 知识星球：https://t.co/kyzMiDmFWb
PSA: X webview can't do redirect, so if your product homepage is a redirect (e.g., to an app store link), fix it so you don’t lose juicy traffic! I've seen a few links here that just go to an empty page.
Cc @hieudinh_
Spent an hour trying to figure out where several hundred extra cuda syncs were coming from... seriously torch... // Difference with the python version: unlike the python version, even when // skipping the finiteness checks (error_if_nonfinite = false), this function // will introduce a device <=> CPU synchronization (for devices where that makes // sense!) in order to return a CPU-side `double`. This C++ version therefore // cannot be run fully asynchronously w.r.t. the device of the gradients.
It was syncing for absolutely no reason... fixed here