Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Jun 1, 2024

• Merge pull request karpathy#495 from ChrisDryden/shared_memory

  Show description for a26041a
  karpathy

  authored
  a26041a

  Copy full SHA for a26041a

• Merge pull request karpathy#509 from ngc92/inplace

  Show description for 9cf8c2f
  karpathy

  authored
  9cf8c2f

  Copy full SHA for 9cf8c2f

• Merge pull request karpathy#511 from gordicaleksa/gordicaleksa_fix_docs

  Show description for bd0f036
  karpathy

  authored
  bd0f036

  Copy full SHA for bd0f036

• Update compile cmd in the dev/cuda README

  gordicaleksa

  committed
  8eafd40

  Copy full SHA for 8eafd40

• rename gelu_backward to gelu_backward_inplace, and made it obvious in the code that this is an inplace operation

  ngc92

  committed
  761fb2c

  Copy full SHA for 761fb2c

• added missing checks

  ngc92

  committed
  cc4bed0

  Copy full SHA for cc4bed0

Commits on May 31, 2024

• fix recompute bug

  karpathy

  committed
  9c8edaf

  Copy full SHA for 9c8edaf

• fix bug in resumption of dataloader, have to reload shard

  karpathy

  committed
  0bc24c3

  Copy full SHA for 0bc24c3

• fix argparse comment too

  karpathy

  committed
  1de3bef

  Copy full SHA for 1de3bef

• Merge branch 'ChrisDryden-layernorm_backwards_mem_saving'

  karpathy

  committed
  60a1037

  Copy full SHA for 60a1037

• make things a bit more consistent, keep recompute 1 as default, and add a comment

  karpathy

  committed
  f4874a9

  Copy full SHA for f4874a9

• Merge branch 'layernorm_backwards_mem_saving' of https://github.com/ChrisDryden/llm.c into ChrisDryden-layernorm_backwards_mem_saving

  karpathy

  committed
  629a240

  Copy full SHA for 629a240

Commits on May 30, 2024

• fix bug in readme

  karpathy

  committed
  ed37d92

  Copy full SHA for ed37d92

• refactor the README file because it was getting crazy

  karpathy

  committed
  b48224c

  Copy full SHA for b48224c

• more docs on scripts

  karpathy

  committed
  0fddf06

  Copy full SHA for 0fddf06

• also move pyrun script to scripts/

  karpathy

  committed
  fe36e40

  Copy full SHA for fe36e40

• code refactor, introduce llmc dir and scripts dir

  Show description for d98d3ed
  karpathy

  authored
  d98d3ed

  Copy full SHA for d98d3ed

• delete these files, they are now in scripts/ folder and better

  karpathy

  committed
  22d1607

  Copy full SHA for 22d1607

• adjust readme to new scripts/ dir

  karpathy

  committed
  fc40ffc

  Copy full SHA for fc40ffc

• scripts folder

  karpathy

  committed
  a0d5dbb

  Copy full SHA for a0d5dbb

• more refactor

  karpathy

  committed
  35aa110

  Copy full SHA for 35aa110

• Merge branch 'master' into feature/libdir

  karpathy

  committed
  f8f9e49

  Copy full SHA for f8f9e49

• further hide the c pytorch bridge logic

  karpathy

  committed
  5f54760

  Copy full SHA for 5f54760

• bugfix don't train on val batch

  karpathy

  committed
  ca27eb8

  Copy full SHA for ca27eb8

• modified the recalculation parameter to be based off of 2 instead of 1

  ChrisDryden

  committed
  59432fe

  Copy full SHA for 59432fe

• Rearranged layernorm forward for 1 and 2 and used layer variables instead. Validated with test file

  ChrisDryden

  committed
  e5719e1

  Copy full SHA for e5719e1

• First step in creating option to recalculate layernorm activations in backwards pass

  ChrisDryden

  committed
  95b1fad

  Copy full SHA for 95b1fad

• Merge pull request karpathy#496 from ChrisDryden/patch-7

  Show description for 0e22982
  karpathy

  authored
  0e22982

  Copy full SHA for 0e22982

• fix default number of gpus from 4 to 8 for consistency with run124M.sh

  karpathy

  committed
  862c957

  Copy full SHA for 862c957

• Merge pull request karpathy#494 from karpathy/feature/pytorch_parity

  Show description for ed805f6
  karpathy

  authored
  ed805f6

  Copy full SHA for ed805f6

• small comments for pyrun124M.sh

  karpathy

  committed
  5959851

  Copy full SHA for 5959851

• fix bug with advancing position in dataloader

  karpathy

  committed
  9b935a8

  Copy full SHA for 9b935a8

• Nit: duplicated line in header of project

  ChrisDryden

  authored
  3bca31f

  Copy full SHA for 3bca31f

• Removed unnecesary shared memory due to blockreduce using static defined shared memory

  ChrisDryden

  committed
  5450632

  Copy full SHA for 5450632

• more feature parity work, i think it's quite close now

  karpathy

  committed
  555d075

  Copy full SHA for 555d075

Pagination

Previous

Next