Commits

Branch selector

 main

User selector

definitelynotmcarilli

Datepicker

All time

Commit History

Commits on Jul 10, 2020

• Small clarification of torch.cuda.amp multi-model example (pytorch#41203)

  Show description for d927aee
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  d927aee

  Copy full SHA for d927aee

Commits on Jul 8, 2020

• Added index_put to promotelist (pytorch#41035)

  Show description for 0911c1e
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  0911c1e

  Copy full SHA for 0911c1e

Commits on Jun 24, 2020

• Allow torch.cuda.amp.GradScaler to support sparse gradients (pytorch#36786)

  Show description for b4ccdef
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  b4ccdef

  Copy full SHA for b4ccdef

Commits on Jun 23, 2020

• Make custom_fwd a no-op when not executed under autocast (pytorch#36171)

  Show description for 3b040c4
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  3b040c4

  Copy full SHA for 3b040c4

• [RELAND2] Change AccumulateGrad to yield .grads that match weights' memory layout (pytorch#40358)

  Show description for 8066fba
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  8066fba

  Copy full SHA for 8066fba

Commits on Jun 17, 2020

• [RELAND] Change AccumulateGrad to yield .grads that match weights' memory layout (pytorch#40129)

  Show description for 1ec8ece
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  1ec8ece

  Copy full SHA for 1ec8ece

Commits on Jun 16, 2020

• Change AccumulateGrad to yield .grads that match weights' memory layout (pytorch#34904)

  Show description for 2beb969
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  2beb969

  Copy full SHA for 2beb969

Commits on May 14, 2020

• Allow GradScaler to be pickled (pytorch#38296)

  Show description for 25f9185
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  25f9185

  Copy full SHA for 25f9185

Commits on May 7, 2020

• Give at::cuda::blas::gemv<at::Half> parity with <float> and <double>. Nature is healing. (pytorch#37569)

  Show description for 35693e9
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  35693e9

  Copy full SHA for 35693e9

Commits on Apr 17, 2020

• Amp gradient accumulation example (pytorch#36601)

  Show description for e6bc34f
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  e6bc34f

  Copy full SHA for e6bc34f

Commits on Mar 24, 2020

• [RELAND2] Eager autocasting, out-of-place ops only (with MSVC 2017 fix) (pytorch#35102)

  Show description for 0f0271e
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  0f0271e

  Copy full SHA for 0f0271e

Commits on Mar 20, 2020

• [RELAND] Eager autocasting, out-of-place ops only (with MSVC 2017 fix) (pytorch#35011)

  Show description for 991b972
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  991b972

  Copy full SHA for 991b972

Commits on Mar 18, 2020

• Eager autocasting, out-of-place ops only (pytorch#32140)

  Show description for aaa8f02
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  aaa8f02

  Copy full SHA for aaa8f02

Commits on Feb 28, 2020

• Formatting changes for gradient scaling (pytorch#33832)

  Show description for a726827
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  a726827

  Copy full SHA for a726827

Commits on Feb 26, 2020

• [WIP] Reanimate gradient scaling API with original scale update heuristic (pytorch#33366)

  Show description for fc6a153
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  fc6a153

  Copy full SHA for fc6a153

Commits on Feb 24, 2020

• Bounds checking for functor execution in vectorized/unrolled kernels (pytorch#33642)

  Show description for e1bddbb
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  e1bddbb

  Copy full SHA for e1bddbb

Commits on Feb 13, 2020

• Gradient scaling API (pytorch#26512)

  Show description for 40246fa
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  40246fa

  Copy full SHA for 40246fa

Commits on Jan 23, 2020

• Fix race condition for to() backward that spans devices (pytorch#31930)

  Show description for 4bdfc71
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  4bdfc71

  Copy full SHA for 4bdfc71

Commits on Nov 21, 2019

• TensorTypeId-based non-RAII setter/getter for LocalTensorTypeSet (pytorch#30113)

  Show description for e05e90c
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  e05e90c

  Copy full SHA for e05e90c

Commits on Nov 1, 2019

• Making torch/csrc/cuda nccl usage safe for nccl 2.5 (pytorch#29014)

  Show description for 314066b
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  314066b

  Copy full SHA for 314066b

Commits on Jun 14, 2019

• Fix with emit_nvtx, also allow shape information to appear in nvtx ranges. (pytorch#21691)

  Show description for 2d5ce51
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  2d5ce51

  Copy full SHA for 2d5ce51

Commits on Feb 11, 2019

• Allow dataloader to accept a custom memory pinning function (pytorch#16743)

  Show description for 0742874
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  0742874

  Copy full SHA for 0742874

Commits on Jan 4, 2019

• Cudnn Handle Pool 3: At Wit's End (pytorch#15668)

  Show description for e313f1a
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  e313f1a

  Copy full SHA for e313f1a

Commits on Dec 13, 2018

• Use a pool of per-thread cudnn handles for each device, updated (pytorch#15080)

  Show description for ca4358c
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  ca4358c

  Copy full SHA for ca4358c

Commits on Dec 11, 2018

• Stashing checkpointing RNG states based on devices of arg tensors (pytorch#14518)

  Show description for 5d3a347
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  5d3a347

  Copy full SHA for 5d3a347

Commits on Nov 30, 2018

• Accumulate grad fix (pytorch#14587)

  Show description for edb3ddf
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  edb3ddf

  Copy full SHA for edb3ddf

Commits on Nov 29, 2018

• Added launch bounds in VolumetricConvolution.cu (pytorch#14564)

  Show description for a2d8e84
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  a2d8e84

  Copy full SHA for a2d8e84

Commits on Nov 23, 2018

• Allow dataloader to accept a custom memory pinning function (pytorch#14171)

  Show description for 7557a99
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  7557a99

  Copy full SHA for 7557a99

• Option to preserve bitwise accuracy of gradient checkpointed vs non-checkpointed dropout (pytorch#14253)

  Show description for c36156e
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  c36156e

  Copy full SHA for c36156e

Commits on Nov 13, 2018

• Minor fix to reenable nvtx sequence numbers for the forward methods of custom (Python) autograd functions (pytorch#13876)

  Show description for 0d2762e
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  0d2762e

  Copy full SHA for 0d2762e

Commits on Oct 29, 2018

• Show demangled names on nvtx ranges (pytorch#13154)

  Show description for 74ac86d
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  74ac86d

  Copy full SHA for 74ac86d

Commits on Sep 14, 2018

• Augment emit_nvtx to help connect backward-pass Function apply calls with their corresponding forward pass ops (pytorch#10881)

  Show description for 0c26488
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  0c26488

  Copy full SHA for 0c26488

Commits on Sep 12, 2018

• Fused weightnorm for ATen (pytorch#10842)

  Show description for a3036b3
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  a3036b3

  Copy full SHA for a3036b3

Commits on Aug 28, 2018

• Future-proofing embedding.py against heuristic changes (pytorch#10959)

  Show description for 579bc43
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  579bc43

  Copy full SHA for 579bc43

Commits on Jul 13, 2018

• Improving THCReduce.cuh's performance on latency-bound non-contiguous reductions (pytorch#9214)

  Show description for cc9dcdf
  

  definitelynotmcarilli

  authored and

  facebook-github-bot

  committed
  cc9dcdf

  Copy full SHA for cc9dcdf