Vijay
1.5K posts
@__tensorcore__
Systems and GPU Performance Mechanic - TBD Ex. CUTLASS 3.x / 4.x etc
🚨 A junior at Jane Street reportedly landed a $220K–$600K role because he used AI to analyze trillions of data points faster than most teams ever could. In this 1-hour lecture, he breaks down the exact system behind it: • how he researches massive datasets • how AI finds patterns humans miss • how his machine turns raw data into decisions • how you can apply the same thinking yourself Skip Netflix tonight. Watch this instead. One hour could completely change how you think about research, AI, and opportunity.
We made Muon run up to 2x faster for free! Introducing Gram Newton-Schulz: a mathematically equivalent but computationally faster Newton-Schulz algorithm for polar decomposition. Gram Newton-Schulz rewrites Newton-Schulz such that instead of iterating on the expensive rectangular X matrix, we iterate on the small, square, symmetric XX^T Gram matrix to reduce FLOPs. This allows us to make more use of fast symmetric GEMM kernels on Hopper and Blackwell, halving the FLOPs of each of those GEMMs. Gram Newton-Schulz is a drop-in replacement of Newton-Schulz for your Muon use case: we see validation perplexity preserved within 0.01, and share our (long!) journey stabilizing this algorithm and ensuring that training quality is preserved above all else. This was a super fun project with @noahamsel, @berlinchen, and @tri_dao that spanned theory, numerical analysis, and ML systems! Blog and codebase linked below 🧵
The newest model in the Mamba series is finally here 🐍 Hybrid models have become increasingly popular, raising the importance of designing the next generation of linear models. We've introduced several SSM-centric ideas to significantly increase Mamba-2's modeling capabilities without compromising on speed. The resulting Mamba-3 model has noticeable performance gains over the most popular previous linear models (such as Mamba-2 and Gated DeltaNet) at all sizes. This is the first Mamba that was student led: all credit to @aakash_lahoti @kevinyli_ @_berlinchen @caitWW9, and of course @tri_dao!
Asymmetric hardware scaling is here. Blackwell tensor cores are now so fast, exp2 and shared memory are the wall. FlashAttention-4 changes the algorithm & pipeline so that softmax & SMEM bandwidth no longer dictate speed. Attn reaches ~1600 TFLOPs, pretty much at matmul speed! joint work w/ Markus Hoehnerbach, Jay Shah(@ultraproduct), Timmy Liu, Vijay Thakkar (@__tensorcore__ ), Tri Dao (@tri_dao) 1/
