FlashMLA

Headlines · 3

1. 1d
   ForgeKernel reportedly beats DeepSeek FlashMLA by 9.8%

   QuantumBit reported that ModelBest’s unattended ForgeKernel agent generated a FlashMLA Prefill kernel for an NVIDIA B200 without consulting DeepSeek’s implementation. The livestreamed run used 42 iterations over 22.5 hours, with the final kernel taking 472 microseconds versus 518 microseconds for FlashMLA, a reported 9.8% improvement. QuantumBit also said ForgeKernel subsequently became open source through the Intelligent Evolution Laboratory and the OpenBMB community.

   量子位（原生 RSS） · 🔥 4
2. 7d
   SGLang Reports Matching or Beating DeepInfer

   DeepSeek open-sourced DeepInfer, the inference engine behind its own service, along with related kernels. In project-run tests using DeepInfer’s benchmarks and the same hardware, SGLang reached up to 1.04x prefill throughput on four GB300 or four B300 GPUs and cut decode step time by as much as 7.4% on 64 GB300 GPUs, according to the SGLang report. A 36氪 report said vLLM also approached or partly exceeded DeepInfer under its disclosed configurations. The comparisons used project-reported results and differing workloads and measurement methods.

   LMSYS Blog · 🔥 0
3. Oct 7
   DeepSeek open-sources Ascend ports of DeepGEMM and DeepEP kernels

   DeepSeek has published two new repositories, DeepGEMM-Ascend for matrix multiplication and DeepEP-Ascend for the all-to-all communication used by mixture-of-experts models, alongside Ascend-targeted updates to TileLang, TileKernels, DeepSelect and FlashMLA. The DeepGEMM changelog dates the Ascend release to 2026.09.30; DeepGEMM itself is a tensor-core kernel library covering FP8, FP4 and BF16 GEMMs, fused MoE and MQA scoring, with all kernels compiled at runtime through DeepJIT. SemiAnalysis reported that at the time of writing only TileLang was usable, because DeepSeek requires CANN 9.2.0 and a C++20 compiler that the beta toolkit did not provide, while TechRadar said DeepGEMM-Ascend needs Ascend 950 series silicon and CANN 9.20. The README's claim that performance matches or exceeds expert-tuned libraries across matrix shapes is the maintainers' own and has not been independently measured.

   GitHub Trending（每日） · 🔥 0

Experience and discussion from the community

Share my FlashMLA experienceAsk about FlashMLA