Commits

Branch selector

 master

User selector

lancerts

Datepicker

All time

Commit History

Commits on May 27, 2024

• amend

  lancerts

  committed
  47c670b

  Copy full SHA for 47c670b

• spell explicitly uint to unsigned int

  lancerts

  committed
  c4c985c

  Copy full SHA for c4c985c

• fix the issue Mismatch of dweight at layernorm_backward.cu

  lancerts

  committed
  f27ca4d

  Copy full SHA for f27ca4d

Commits on May 20, 2024

• update the utils function and assert

  lancerts

  committed
  2b0667a

  Copy full SHA for 2b0667a

• fix the unsupported block_size

  lancerts

  committed
  6348d41

  Copy full SHA for 6348d41

Commits on May 10, 2024

• update the -lcublas -lcublasLt flag in the comment

  lancerts

  committed
  5c90845

  Copy full SHA for 5c90845

Commits on May 5, 2024

• Minor update on the code

  lancerts

  committed
  bbfe8c9

  Copy full SHA for bbfe8c9

Commits on May 4, 2024

• minor fix of the include

  lancerts

  committed
  03b7323

  Copy full SHA for 03b7323

Commits on May 3, 2024

• move gelu_backward to backward block

  lancerts

  committed
  c81adeb

  Copy full SHA for c81adeb

Commits on Apr 28, 2024

• amend the float4 kernel

  lancerts

  committed
  1c7d23a

  Copy full SHA for 1c7d23a

• amend the float4 kernel

  lancerts

  committed
  61c5c05

  Copy full SHA for 61c5c05

• Include the float4 kernel

  lancerts

  committed
  cfccd82

  Copy full SHA for cfccd82

Commits on Apr 19, 2024

• fix the constness and the warning C4305: 'initializing': truncation from 'double' to 'float'

  lancerts

  committed
  3cb993b

  Copy full SHA for 3cb993b

• fix the 0.0f initialization for double

  lancerts

  committed
  0c92ea9

  Copy full SHA for 0c92ea9

• fix the 0.0f initialization for double

  lancerts

  committed
  e991fe1

  Copy full SHA for e991fe1

Commits on Apr 18, 2024

• fix the bug that yields cpu, gpu results mismatch

  lancerts

  committed
  a1abcbc

  Copy full SHA for a1abcbc

• fix the bug that yields cpu, gpu results mismatch

  lancerts

  committed
  ac055f7

  Copy full SHA for ac055f7

Commits on Apr 17, 2024

• Constness fix layernorm_backward.cu

  lancerts

  committed
  c327923

  Copy full SHA for c327923

• Constness fix layernorm_backward.cu

  lancerts

  committed
  81aa596

  Copy full SHA for 81aa596

Commits on Apr 16, 2024

• fix the backward dbias precision issue

  lancerts

  committed
  6ff4ad6

  Copy full SHA for 6ff4ad6

Commits on Apr 11, 2024

• Include a parallel kernal of softmax_online

  lancerts

  committed
  69d0e5b

  Copy full SHA for 69d0e5b

• Include a parallel kernal of softmax_online

  lancerts

  committed
  ca1e5ee

  Copy full SHA for ca1e5ee

• Improve the efficiency

  lancerts

  committed
  d6bcc12

  Copy full SHA for d6bcc12

• Improve the efficiency

  lancerts

  committed
  00010c7

  Copy full SHA for 00010c7

• Include the online softmax CPU code (from the paper "Online normalizer calculation for softmax") and native port to GPU kernel.

  Show description for f524a90
  lancerts

  committed
  f524a90

  Copy full SHA for f524a90

Commits on Apr 10, 2024

• fix error in small typos in matmul_forward.cu

  lancerts

  committed
  68d2ce0

  Copy full SHA for 68d2ce0

Commits on Apr 9, 2024

• fix the consistency of the transpose notation

  lancerts

  committed
  fbd679a

  Copy full SHA for fbd679a

• amend

  lancerts

  committed
  3a4d783

  Copy full SHA for 3a4d783

• fix error: identifier M_PI is undefined

  lancerts

  committed
  73c9291

  Copy full SHA for 73c9291

• Include the pytorch layer_norm.cpp and layer_norm_kernel.cu code pointer for people who are interested in details

  lancerts

  committed
  eecb234

  Copy full SHA for eecb234