1. Top
  2. transformer-shortest-paths
  3. Commits
transformer-shortest-paths

Experimentally evaluating transformers' generalization on a synthetic task (mirror)

Commits at 78b208c42761ed0f43c11fca9d20d3ef1d5ca85f

  1. 78b208c4 add stuff SIPB authored at SIPB comitted at
  2. 95f0dcac Commit everything SIPB authored at SIPB comitted at
  3. 0e78a419 adding some stuff SIPB authored at SIPB comitted at
  4. ae08546a Add tune loss too SIPB authored at SIPB comitted at
  5. 7f3f049e Add nearly finished code SIPB authored at SIPB comitted at
  6. ad1458c0 Commit more stuff SIPB authored at SIPB comitted at
  7. 6ae42b74 Commit everything SIPB authored at SIPB comitted at
  8. f54a040e Oh oops delete those files too SIPB authored at SIPB comitted at
  9. 889c0de5 Move *.dot to img too SIPB authored at SIPB comitted at
  10. 1ea92f93 More work on blog.md, move images to img folder SIPB authored at SIPB comitted at
  11. 77ad0081 Add super good code SIPB authored at SIPB comitted at
  12. 7c101c46 update SIPB authored at SIPB comitted at
  13. 74629688 Latest blog post and graphs SIPB authored at SIPB comitted at
  14. a24288e2 Latest copy of blog post and insane TSP SIPB authored at SIPB comitted at
  15. 58e39027 Get train err down to .35, add model and loss file SIPB authored at SIPB comitted at
  16. 55f2c364 Add first draft of blog post SIPB authored at SIPB comitted at
  17. 3982e8fa Make the target embedding learnable SIPB authored at SIPB comitted at
  18. c8c1abdd Better plots SIPB authored at SIPB comitted at
  19. 072c87a3 New embeddings and readout SIPB authored at SIPB comitted at
  20. 68887a78 Decrease num layers SIPB authored at SIPB comitted at
  21. eab86fac MORE LAYERS SIPB authored at SIPB comitted at
  22. e953b375 Use bfloat16 SIPB authored at SIPB comitted at
  23. e460b0ae idk what this does but it's the latest version of TSP SIPB authored at SIPB comitted at
  24. 5654942c Generate data on the fly SIPB authored at SIPB comitted at
  25. 1fe0e023 Minor README tweaks SIPB authored at SIPB comitted at
  26. d3e7c909 updates Alek Westover authored at Alek Westover comitted at
  27. 461d3543 Delete unused images dir SIPB authored at SIPB comitted at
  28. 11d377f6 Condense proposal into 1 page sipb authored at sipb comitted at
  29. 26b5a9d8 wrote project proposal @anthowan Anthony can you make sure it looks good? Alek Westover authored at Alek Westover comitted at
  30. 49059938 Tasks 1 through 4 SIPB authored at SIPB comitted at