區塊先生 🐡 ⚠️ (rock #58)
@mrblock
investor & advisor at @CurveFinance @lighter_xyz & Puffer 🎍I make Crypto videos 平時就是做做Crypto影片和天使投資
加入 May 2019
4.1K 正在关注 99.8K 粉丝
又一家AI對手出來了！
SubQ 推出了首款 frontier LLM，採用 fully sub-quadratic sparse-attention (SSA) 架構，context window 達 1,200 萬 tokens。
重點數據： •
計算量約為傳統 quadratic Transformer 的 1/1000 • 1M token prefill 比 FlashAttention 快 52 倍 • 成本不到 Claude Opus 的 5%
Benchmark：
• SWE-Bench Verified：81.8%
• RULER @128K：95%
早期存取與 SubQ Code agent 已開放：
显示更多
Introducing SubQ - a major breakthrough in LLM intelligence.
It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA),
And the first frontier model with a 12 million token context window which is:
- 52x faster than FlashAttention at 1MM tokens
- Less than 5% the cost of Opus
Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention).
Only a small fraction actually matter.
@subquadratic finds and focuses only on the ones that do.
That's nearly 1,000x less compute and a new way for LLMs to scale.
显示更多
