Kernel Design Agents optimize Kimi Delta Attention kernels, up to 2.96x speedup
songhan_mit · x · 2026-09-29
Ligeng Zhu announces KDA(gent)-v0.6: Kernel Design Agents optimizing Kimi Delta Attention (KDA) kernels with up to 2.96x speedup — the project acronym happens to collide with KDA, so the team literally used KDA to optimize KDA.
- Iterative LLM collaboration: a Humanize2 Flow running multi-round evolution with Flame Chase, GPT-5.6-Sol and Fable-5, plus automated workspace cleanup, greatly broadening the operator search space
- Multilingual domain skill stack: full support for CuTe-DSL, CUDA C++, and the new agent-native CAKE IR and TIRx, with custom diagnostic tooling
- A showcase of a full LLM-agent workflow for automatic GPU kernel authoring and optimization
More from Infra
- Sarvam AI Launches Token Factory: India-Hosted Inference for Open Models — itsOmSarraf_ · 2026-11-16
- BIS Issues Interim Final Rule Addressing Remote Access to Advanced AI Chips — apples_jimmy · 2026-11-16
- Jailbroken PS4 runs a 9B open-source model at up to 43 tok/s by talking directly to its 2013-era GPU — AIFlow_ML · 2026-11-16
- Mobile AI Factory: agents auto-build on-device SLMs with 19x less GPU time — pmttyji · 2026-11-16
- Stratechery's Fictional Obituary: Imagining Jensen Huang's 2026 Exit From Nvidia — Gregory_C_Allen · 2026-11-16
- vLLM serves DeepSeek V4.1-Flash at 954K tokens/s per GPU, $38.96/GPU-hour on GB300 NVL72 — bookwormengr · 2026-11-16