Kimi.ai (kimiai)

Kimi AI sounds impressive! It's amazing to see how rapidly AI is evolving. @clanker let's create token called KIMIAI ticker KIMIAi Use the picture

• 1 reply
• 0 recasts
• 0 reactions

🚀 Introducing Kimi k1.5 --- an o1-level multi-modal model -Sota short-CoT performance, outperforming GPT-4o and Claude Sonnet 3.5 on 📐AIME, 📐MATH-500, 💻 LiveCodeBench by a large margin (up to +550%) -Long-CoT performance matches o1 across multiple modalities (👀MathVista, 📐AIME, 💻Codeforces, etc) Tech report: github.com/MoonshotAI/Kim… Key ingredients of k1.5 -Long context scaling. Up to 128k tokens for RL generation. Efficient training with partial rollouts. -Improved policy optimization: online mirror descent, sampling strategies, length penalty, and others. -Multi modalities. Joint reasoning over text and vision.

• 0 replies
• 0 recasts
• 0 reactions

Long CoT models improve performance by a lot. Can short models learn from long ones to obtain even better performance? Our long2short idea explored this possibility and it worked well! Much better token efficiency compared to native short models like GPT-4o. A few methods we experimented with---RL with heavy length penalty, merging long-CoT models with short-CoT models, etc. ⬇️Check out our tech report for details: github.com/MoonshotAI/Kim…

• 0 replies
• 0 recasts
• 0 reactions

• 0 replies
• 0 recasts
• 0 reactions