Code

History

Commits on move/encoder_kernel to Makefile35+
  1. moved device utilities into a separate file, and follow-up cleanupErik Schultheis committed 2 years ago67502c7
  2. small more fixesAndrej Karpathy committed 2 years ago020b15a
  3. attempt to remove PATH_SEP from MakefileAndrej Karpathy committed 2 years ago6434009
  4. attempt to move cudnn attention to the llmc directoryAndrej Karpathy committed 2 years ago41dc65a
  5. Check if file exists using platform specific commands & Add *.o to gitignoreJun Zhang committed 2 years ago9e64531
  6. Fix detection of cudnn-frontend in '.' on LinuxMarco van Zwetselaar committed 2 years ago6da5e63
  7. Adds cuDNN install instructions for a smaller but sufficient packageJoe Halabi committed 2 years ago69aa64c
  8. Rename cudnn_att.cu to cudnn_att.cpp to speed up compilation.Anerudhan committed 2 years ago0ce5fcf
  9. Adding two directory search for cuDNN frontend filesRoss Wheeler committed 2 years ago134f4c7
  10. Merge branch 'karpathy:master' into cudnn_makefile_changesrosslwheeler committed 2 years agodedd86c
  11. Adding CI check to disable auto-detectRoss Wheeler committed 2 years ago0c4908d
  12. Auto-detect GPU capabilityRoss Wheeler committed 2 years ago2bfd2b6
  13. Minor fixes for Makefile for cudnn and windows supportRoss Wheeler committed 2 years ago2d4e5fd
  14. don't compile/link cudnn if not asked for itErik Schultheis committed 2 years agob087b9c
  15. fixed up test and profile targetsErik Schultheis committed 2 years agoaa5bb25
  16. first attempt at moving cudnn out of the main file for faster compilesErik Schultheis committed 2 years ago67a82a6
  17. small fixes for profilingAndrej Karpathy committed 2 years agoe780b56
  18. add readme and search for cudd in home dirAndrej Karpathy committed 2 years ago69d40eb
  19. updates to makefile: docs, override, etcAndrej Karpathy committed 2 years ago3564065
  20. round 1 of small fixes for cudnn supportAndrej Karpathy committed 2 years ago04718fb
  21. Add makefile changes for cuDNNademeure committed 2 years ago7beff72
  22. fix profile gpt2cu but note it currently needs the NO_MULTI_GPU=1 switch i think when calling it with make, todo fix later how we add the mpi flagsAndrej Karpathy committed 2 years ago906d22f
  23. Adding back PFLAGSRoss Wheeler committed 2 years agofcef97d
  24. Updating Makefile with latest changesRoss Wheeler committed 2 years agoc9ab8ee
  25. Merge pull request #279 from Ricardicus/prec-ciAndrej committed 2 years ago18b41b4
  26. add small comment on -t=0Andrej Karpathy committed 2 years ago49228b0
  27. Updating the CI to build different precisionsRickard Hallerbäck committed 2 years agob522333
  28. Merge pull request #269 from ChrisDryden/patch-3Andrej committed 2 years ago10aa24e
  29. profile and test only use bf16. but the train script can be run with fp32 or bf16 or fp16. fp16 will error, thoughAndrej Karpathy committed 2 years ago9d70d9a
  30. allow user to make different precisions, add prints and error handling around precisionsAndrej Karpathy committed 2 years ago4f7d8d9
  31. Enable multithreading in nvccChris Dryden committed 2 years ago2954d90
  32. fix bug with all_hostsname_hashes missing, re-arrange code a bit, make sure we only print on rank0Andrej Karpathy committed 2 years agoa00a3a8
  33. more defensive check for openmpi install, and an install commandAndrej Karpathy committed 2 years ago2e17140
  34. [Multi-GPU] llm.c now runs on multiple GPUs with NCCLPetr Zhizhin committed 2 years agod49e830
  35. updates to readme, and introduce the test fp32 cuda file tooAndrej Karpathy committed 2 years agoed6387a