write-ups >

  1. SplitCompute and working with Tensors
  2. "Threads cannot be implemented as a library"
  3. Talking to GPUs
  4. The Fortran comeback?
  5. Offline RL Simulation
  6. Interaction Nets
  7. One more ML framework will fix all problems
  8. Quantized LoRA
  9. Contravariance and Profunctors
  10. McKelvey-Schofield chaos theorem
  11. Virtual Water
  12. Forward mode Autodiff and Dual Numbers
  13. Beware of call overheads
  14. Fast array summation with CUDA
  15. Volume of a hypersphere
  16. mmap() and the confusion with LLaMa's sparsity
  17. How did transformers get this good?