Hackernews posts about MoE

1. 631
   

   newStrata: Qwen3.8-Flash-Next (125B Moe) on a 8GB+ Nvidia GPUgithub.com

   ▲ 5simonpure7ddiscusssavesaved
2. 632
   

   newStepFun teases Step 5 Preview, a 600B/27B MoE, open weights coming Oct 15twitter.com

   ▲ 3WavyPeng13d💬 1savesaved
3. 633
   

   newSSD-Llama: SSD-Native Inference for Trillion-Parameter Moe on a Consumer PCarxiv.org

   ▲ 3AlmostCosmo7917ddiscusssavesaved
4. 634
   

   newShow HN: Llama.cpp fork with 2-4x multiGPU speed for MoE models bigger than VRAMgithub.com

   ▲ 2neuralll10d💬 3savesaved
5. 635
   

   newI joined the 'anti-chatbox' interface movement, and build Moe because of itmoebot.app

   ▲ 2yisrael_g19d💬 1savesaved
6. 636
   

   newShow HN: Moe Routing Atlas – which experts fire in Qwen3.5/3.6 35B-A3Bgslaller.github.io

   ▲ 2gslaller6ddiscusssavesaved
7. 637
   

   newXing4.0-29B-A4B: China Telecom's Moe Model Trained on Ascend NPUswww.mindstudio.ai

   ▲ 2Bluestein15ddiscusssavesaved
8. 638
   

   newShow HN: I made a tiny MoE/Engram viz toolsw-ml-study.github.io

   ▲ 2softwarewright21ddiscusssavesaved
9. 639
   

   newThe Newsroom EP01 – OpenAI Ships GPT-5 to Azure – DeepSeek Open-Sources 236B Moewww.youtube.com

   ▲ 2LebToki23ddiscusssavesaved
10. 640
    

    newEmpirical Research on MoE Safety Mechanismsdealign.ai

    ▲ 2Bluestein24ddiscusssavesaved
11. 641
    

    newExtending Raschka's GPT-2: an MoE trained from scratch on an RTX 3090www.gilesthomas.com

    ▲ 2ibobev24ddiscusssavesaved
12. 642
    

    newDense vs. Moe Models: Active Parameters, Throughput, and When to Choose Eachdeveloper.nvidia.com

    ▲ 1pretext19ddiscusssavesaved
13. 643
    

    newExtending Raschka's GPT-2: an MoE trained from scratch on an RTX 3090www.gilesthomas.com

    ▲ 1gpjt25ddiscusssavesaved
14. 644
    

    newFractal-BLT – Zero-allocation .NET 10 MoE runtime streaming NVMe to GPUgithub.com

    ▲ 1Ctrl_Alt_Haze27ddiscusssavesaved
15. 645
    

    newShow HN: Mini-AGI – Dynamic continual learning model trained on 8GB VRAMgithub.com

    ▲ 277volotat14d💬 81savesaved
16. 646
    

    newAsk HN: What do AI improvements at these labs look like?

    ▲ 2xyzsparetimexyz22ddiscusssavesaved
17. 647
    

    newShow HN: Way to divide parameter space in LLM trainingzenodo.org

    ▲ 1genji9709ddiscusssavesaved
18. 648
    

    newShow HN: PicoLM v1.0-rc2 ("Yura Kana"). Run an LLM on Digital Unixgithub.com

    ▲ 1gabucino12ddiscusssavesaved
19. 649
    

    newReverse engineering my e-scooter and rewriting the firmware in Rustbensimms.moe

    ▲ 452vinhnx25d💬 105savesaved
20. 650
    

    newDon't call yourself an artisanal programmerpurplesyringa.moe

    ▲ 57emschwartz22d💬 65savesaved
21. 651
    

    newVibe Coding Production Kit – a production workflow for AI coding agentsgithub.com

    ▲ 9Moeeryani11d💬 6savesaved
22. 652
    

    newGrieving the Loss of Detailspurplesyringa.moe

    ▲ 6ibobev11ddiscusssavesaved
23. 653
    

    newWrite While Learningpurplesyringa.moe

    ▲ 6dalvrosa16ddiscusssavesaved
24. 654
    

    newGrieving the Loss of Detailspurplesyringa.moe

    ▲ 4ibobev10d💬 2savesaved
25. 655
    

    newWrite While Learningpurplesyringa.moe

    ▲ 3meetpateltech16ddiscusssavesaved
26. 656
    

    newSqueezing rust onto more things, this time: Flashlightsbensimms.moe

    ▲ 3Bluestein22ddiscusssavesaved
27. 657
    

    newEvergardenevergarden.moe

    ▲ 3birdculture24ddiscusssavesaved
28. 658
    

    newWe Spoke with Entrepreneurs Living the "7 Vibe 3" Lifestylecadence.moe

    ▲ 2xyzsparetimexyz25d💬 3savesaved
29. 659
    

    newEvergardenevergarden.moe

    ▲ 2birdculture14ddiscusssavesaved
30. 660
    

    newReverse engineering my E-Scooter and rewriting it in Rustbensimms.moe

    ▲ 2nitros27ddiscusssavesaved

· edit