Commits

Breadcrumbs

History for

onmaster

User selector

All users

Datepicker

All time

Commit History

Commits on Jul 18, 2024

• allocate_weights utility function

  Show description for 0bcb3e5
  ngc92

  committed
  0bcb3e5

• allocate_state utility function

  ngc92

  committed
  cde79ff

• consolidate model init

  ngc92

  committed
  b20def9

Commits on Jul 1, 2024

• zero-grad is async and part of backward call

  ngc92

  committed
  486f98e

• Add micro step guard

  gordicaleksa

  committed
  74e5ba7

Commits on Jun 26, 2024

• resolve merge conflicts

  karpathy

  committed
  e12d090

Commits on Jun 25, 2024

• fix up profile

  ngc92

  committed
  4aae1e5

• fix profile

  ngc92

  committed
  2413b0a

• fix training loop in other parts

  karpathy

  committed
  329aee3

Commits on Jun 24, 2024

• Merge pull request #631 from gordicaleksa/minor_refactor2

  Show description for 76b133d
  karpathy

  authored
  76b133d

• Fix arg permutation

  gordicaleksa

  committed
  c420a52

• Fix CI errors

  gordicaleksa

  committed
  a259eea

Commits on Jun 22, 2024

• Rename gpt2 backward

  gordicaleksa

  committed
  71f10e7

Commits on Jun 17, 2024

• added a determinism check

  ngc92

  committed
  4d35bac

Commits on Jun 12, 2024

• fix non-multi-gpu case

  ngc92

  committed
  3686449

Commits on Jun 3, 2024

• refactor part 2, moving stuff into common files so that we can later nicely separate out all the kernels as well

  karpathy

  committed
  804d069

Commits on May 21, 2024

• Fully deterministic encoder backward kernels for train_gpt2.cu

  ademeure

  committed
  b5e75dd

Commits on May 19, 2024

• Merge pull request #434 from ngc92/profile-fix

  Show description for 00c291c
  karpathy

  authored
  00c291c

• initialize multi_gpu_config so profile doesn't crash

  ngc92

  committed
  299ce65

Commits on May 18, 2024

• fixed profile target

  ngc92

  committed
  c3a3b9d

Commits on May 16, 2024

• fixup profiling

  ngc92

  committed
  fbd8f03

Commits on May 5, 2024

• Add FP16 path for atomicStochasticAdd ( remove __bfloat1622float2 to work on older CUDAs) fixes

  ademeure

  committed
  876ab93

• Merge remote-tracking branch 'karpathy/master' into cleanup_may4

  ademeure

  committed
  c261eec

• Merge branch 'separate-compilation' of https://github.com/ngc92/llm.c into ngc92-separate-compilation

  karpathy

  committed
  f2224f2

• Merge branch 'less_idle_more_brrr' of https://github.com/ademeure/llm.c into ademeure-less_idle_more_brrr-3

  karpathy

  committed
  2642ffd

Commits on May 4, 2024

• fix merge conflict

  ademeure

  committed
  3430436

• refactoring & remove unused functions to reduce LOC ( wip profile.py improvements)

  ademeure

  committed
  bfb9c51

• fixed up test and profile targets

  ngc92

  committed
  aa5bb25

Commits on May 3, 2024

• Set cuBLAS(Lt) to use the new main_stream

  ademeure

  committed
  a19fcec

• small fixes for profiling

  karpathy

  committed
  e780b56

• super optimised matmul_bias cuda streams for better parallelism fix fused_classifier cache hint remove remaining cooperative groups

  ademeure

  committed
  b581810

• add CUDA streams (no parallelism yet) fix profiling init new "blockReduce" function

  ademeure

  committed
  f456a75

Commits on May 1, 2024

• and also profile lol

  karpathy

  committed
  8aa9a9e

• Add missing cudnnDestroy() calls

  ademeure

  committed
  1147983

• Fixed profile_gpt2.cu/test_gpt2.cu for cuDNN

  ademeure

  committed
  d9e7a0a

Pagination

PreviousNext