Commits

Branch selector

 a8097ee

User selector

All users

Datepicker

All time

Commit History

Commits on Mar 24, 2021

• add approx_sgd/rowwise_adagrad to setup.py (pytorch#579)

  Show description for a8097ee
  

  jspark1105

  authored and

  facebook-github-bot

  committed
  a8097ee

• Remove Explicit prefetch argument from forward (pytorch#578)

  Show description for 7c0c486
  

  another-pjohnson

  authored and

  facebook-github-bot

  committed
  7c0c486

• Back out "have bf16 quantize kernels use 2d blocks with loop to cover large tensors" (pytorch#572)

  Show description for fc2c734
  

  jianyuh

  authored and

  facebook-github-bot

  committed
  fc2c734

Commits on Mar 23, 2021

• Use more compact syntax for is_uvm_tensor (pytorch#577)

  Show description for 1d7eaf3
  

  ezyang

  authored and

  facebook-github-bot

  committed
  1d7eaf3

• use JIT'ed approx rowwise sparse adagrad on CPU (pytorch#574)

  Show description for a2b58df
  

  jspark1105

  authored and

  facebook-github-bot

  committed
  a2b58df

• add grad stride option to fused rowwise sparse adagrad (pytorch#568)

  Show description for 8998e6f
  

  jspark1105

  authored and

  facebook-github-bot

  committed
  8998e6f

• make ref code deterministic in split_table_batched_embeddings_test (pytorch#565)

  Show description for 7f78584
  

  jspark1105

  authored and

  facebook-github-bot

  committed
  7f78584

• Fix OSS build failure due to approx sgd (pytorch#570)

  Show description for 28b130f
  

  jianyuh

  authored and

  facebook-github-bot

  committed
  28b130f

• Move permute_pooled_embs_kernel into fbgemm_gpu (pytorch#569)

  Show description for 3657619
  

  jianyuh

  authored and

  facebook-github-bot

  committed
  3657619

Commits on Mar 22, 2021

• relax error bound of fp16 sparse adagrad (pytorch#567)

  Show description for f3abcb8
  

  jspark1105

  authored and

  facebook-github-bot

  committed
  f3abcb8

• fix approx sgd on cpu (pytorch#566)

  Show description for ffff7a3
  

  jspark1105

  authored and

  facebook-github-bot

  committed
  ffff7a3

• avoid unused variable warning for approx sgd (pytorch#564)

  Show description for 99f3034
  

  jspark1105

  authored and

  facebook-github-bot

  committed
  99f3034

Commits on Mar 20, 2021

• avoid initializing embedding bag output to zeros (pytorch#560)

  Show description for 7060df8
  

  jspark1105

  authored and

  facebook-github-bot

  committed
  7060df8

• array out of bound checking (pytorch#562)

  Show description for 9f0062e
  

  jspark1105

  authored and

  facebook-github-bot

  committed
  9f0062e

• add one more entry to hash_size_cum to easier compute of hash size of each table (pytorch#561)

  Show description for 4ac3a49
  

  jspark1105

  authored and

  facebook-github-bot

  committed
  4ac3a49

• format split_table_batched_embeddings_ops.py (pytorch#563)

  Show description for 2ab5e3f
  

  jspark1105

  authored and

  facebook-github-bot

  committed
  2ab5e3f

Commits on Mar 19, 2021

• correctly handle fp16 table in approx sgd

  Show description for a8155a0
  

  jspark1105

  authored and

  facebook-github-bot

  committed
  a8155a0

• Also include the dependency to FBGEMM CPU SPMDM routine (pytorch#559)

  Show description for e297346
  

  jianyuh

  authored and

  facebook-github-bot

  committed
  e297346

Commits on Mar 18, 2021

• Optimize the performance of recat_embedding_grad_output (pytorch#488)

  Show description for d4f2dcc
  

  jianyuh

  authored and

  facebook-github-bot

  committed
  d4f2dcc

• Add FBGEMM_CPU dependency in setup.py for table batched embedding cpu op (pytorch#556)

  Show description for e4ca7c4
  

  jianyuh

  authored and

  facebook-github-bot

  committed
  e4ca7c4

• Fix the OSS CMakefile build (pytorch#555)

  Show description for 7716e15
  

  jianyuh

  authored and

  facebook-github-bot

  committed
  7716e15

• only test optimizers supported in CPU in test_backward_optimizers (pytorch#557)

  Show description for a064fec
  

  jspark1105

  authored and

  facebook-github-bot

  committed
  a064fec

• Remove "step" from table batched Rowwise Adagrad optimizer state (pytorch#481)

  Show description for 51bc2ec
  

  Wen Zhang (AI Infra)

  authored and

  facebook-github-bot

  committed
  51bc2ec

• Fix OSS build and run issues (pytorch#554)

  Show description for 4ef1ec0
  

  jianyuh

  authored and

  facebook-github-bot

  committed
  4ef1ec0

• Add per-feature precision support in the model level (pytorch#552)

  Show description for 334134c
  

  jianyuh

  authored and

  facebook-github-bot

  committed
  334134c

Commits on Mar 17, 2021

• real sparse fused exact optimizer: SGD and (row-wise) AdaGrad (pytorch#551)

  Show description for 95800eb
  

  jiyuanzFB

  authored and

  facebook-github-bot

  committed
  95800eb

• use fbgemm cpu embedding spmdm jit kernel (pytorch#550)

  Show description for a06cb38
  

  jspark1105

  authored and

  facebook-github-bot

  committed
  a06cb38

• add output stride option to embedding spmdm (pytorch#548)

  Show description for 4ab1eeb
  

  jspark1105

  authored and

  facebook-github-bot

  committed
  4ab1eeb

• Remove dead includes in deeplearning/fbgemm/test (pytorch#534)

  Show description for 88ba128
  

  generatedunixname89002005287564

  authored and

  facebook-github-bot

  committed
  88ba128

Commits on Mar 16, 2021

• Forcing grad_output to be 16 bytes aligned

  Show description for 8ae350d
  

  xing-liu

  authored and

  facebook-github-bot

  committed
  8ae350d

• use 2D parallelism in CPU emb fwd (pytorch#549)

  Show description for a4d3ca7
  

  jspark1105

  authored and

  facebook-github-bot

  committed
  a4d3ca7

• Update asmjit for fbcode and oss (pytorch#538)

  Show description for 5d36b7d
  

  dskhudia

  authored and

  facebook-github-bot

  committed
  5d36b7d

• have bf16 quantize kernels use 2d blocks with loop to cover large tensors (pytorch#541)

  Show description for 2b5c02d
  

  amylittleyang

  authored and

  facebook-github-bot

  committed
  2b5c02d

• Register cpu path table_batched parameters as buffers to enable running on Mast (pytorch#547)

  Show description for bd470d9
  

  Chris Cai

  authored and

  facebook-github-bot

  committed
  bd470d9

Commits on Mar 15, 2021

• use at::acc_type (pytorch#546)

  Show description for 4e7dbd0
  

  jspark1105

  authored and

  facebook-github-bot

  committed
  4e7dbd0

Pagination

PreviousNext