write-ups >

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