Magnus is all grown up 🥹
Started as my baby for wrangling large-scale ML data at ByteDance, now it has its own PVLDB paper on holistic data management for LRMs/LMMs.
Huge congrats to the team pushing it forward!
New Feature: Research & Task 🎉
We’re excited to announce the launch of Research & Task, a new feature that helps users quickly analyze market conditions and turn it into clear next steps.
With Research & Task, users can ask for things like scheduled technical analysis, risk checks, market research workflow, and AI indicators planning in one place. It’s designed to make research faster, easier, and more actionable.
This feature is now available in beta test. Try it out and let us know what you think! 🏆
We push Prefill/Decode disaggregation beyond a single cluster: cross-datacenter + heterogeneous hardware, unlocking the potential for significantly lower cost per token.
This was previously blocked by KV cache transfer overhead. The key enabler is our hybrid model (Kimi Linear), which reduces KV cache size and makes cross-DC PD practical.
Validated on a 20x scaled-up Kimi Linear model:
✅ 1.54× throughput
✅ 64% ↓ P90 TTFT
→ Directly translating into lower token cost.
More in Prefill-as-a-Service: https://t.co/If8fA3t9Og
Magnus is all grown up 🥹
Started as my baby for wrangling large-scale ML data at ByteDance, now it has its own PVLDB paper on holistic data management for LRMs/LMMs.
Huge congrats to the team pushing it forward!