Soumith Chintala

@soumithchintala

@TaikoToeda We will release NVFP4 and MXFP8 checkpoint. The MXFP8 checkpoint has a very marginal 1-2 percentage points gain over the NVFP4 checkpoint over evals (even though model native training is NVFP4), but we don't think its meaningful for any practical use-cases.
打开原帖#511482
  1. Industry

    Elon Musk: @TheFlintFace @IterIntellectus 100GW in space is roughly equal to 800GW on Earth. Moreover, solar on Earth must be paired with massive batteries to continue powering co…
  2. Industry

    Runway: Seedance 2.5 in 4K is here
  3. Industry

    Perplexity: We are publishing new research on our findings from red-teaming SPACE…