Commits

Branch selector

 main

User selector

kvareddy

Datepicker

All time

Commit History

Commits on Jul 31, 2023

• yet another minor format fix

  kvareddy

  committed
  e7cf86e

  Copy full SHA for e7cf86e

• minor fix

  kvareddy

  committed
  be20c6b

  Copy full SHA for be20c6b

Commits on Jul 25, 2023

• Interleave dataset support

  kvareddy

  committed
  622a44b

  Copy full SHA for 622a44b

Commits on Jul 21, 2023

• qkv projection semantics fix

  kvareddy

  committed
  000d291

  Copy full SHA for 000d291

• test regression fix

  kvareddy

  committed
  4e31ee1

  Copy full SHA for 4e31ee1

Commits on Jun 25, 2023

• code refactor + packing text and img in to a single bin file

  kvareddy

  committed
  efd8f78

  Copy full SHA for efd8f78

Commits on Nov 17, 2022

• Data Preprocessing Optimizations

  

  kvareddy

  authored and

  Mike Chrzanowski

  committed
  7fc9611

  Copy full SHA for 7fc9611

Commits on Nov 11, 2022

• Support for variable sequence lengths across micro-batches

  

  kvareddy

  authored and

  jaredcasper

  committed
  c3e688d

  Copy full SHA for c3e688d

Commits on Oct 14, 2022

• inverse_square_root learning param schedule

  kvareddy

  committed
  dbed5e0

  Copy full SHA for dbed5e0

Commits on Oct 11, 2022

• address review comments

  kvareddy

  committed
  69f3249

  Copy full SHA for 69f3249

Commits on Oct 4, 2022

• different encoder/decoder num-layers support

  kvareddy

  committed
  cbf8250

  Copy full SHA for cbf8250

Commits on Oct 3, 2022

• Merge branch 'main' into nmt-main

  kvareddy

  committed
  41276b6

  Copy full SHA for 41276b6

Commits on Aug 8, 2022

• adress review comments

  kvareddy

  committed
  a44360e

  Copy full SHA for a44360e

Commits on Aug 4, 2022

• yttm + BytelevelBPE + setencepeice tokenizer support

  kvareddy

  committed
  45f4ee5

  Copy full SHA for 45f4ee5

Commits on Jul 28, 2022

• support for all mask in fused kernel + avoiding inplace operation in bwd pass

  kvareddy

  committed
  76db958

  Copy full SHA for 76db958

Commits on May 26, 2022

• bug fix for get_data_parallel_src_rank

  kvareddy

  committed
  9ad1944

  Copy full SHA for 9ad1944

Commits on May 24, 2022

• resolved review comments

  kvareddy

  committed
  739cb43

  Copy full SHA for 739cb43

• preallocating global buffer to avoid memory fragmentation

  kvareddy

  committed
  9dc3c42

  Copy full SHA for 9dc3c42

• fix for sequence parallelism in bert pooling

  kvareddy

  committed
  8474e6e

  Copy full SHA for 8474e6e

Commits on May 20, 2022

• avoiding sequence parallelism on the pooler

  kvareddy

  committed
  3f91f09

  Copy full SHA for 3f91f09

• bert regression fixes

  kvareddy

  committed
  6aaafee

  Copy full SHA for 6aaafee

Commits on May 18, 2022

• t5 regression fixes

  kvareddy

  committed
  5d64738

  Copy full SHA for 5d64738

Commits on May 17, 2022

• address review comment

  kvareddy

  committed
  a7a12f8

  Copy full SHA for a7a12f8

• address review comments

  kvareddy

  committed
  d19e8ad

  Copy full SHA for d19e8ad

• address review comments

  kvareddy

  committed
  bbab79f

  Copy full SHA for bbab79f

• address review comments

  kvareddy

  committed
  356eb36

  Copy full SHA for 356eb36

Commits on May 16, 2022

• Merge branch 'main' into sequence_parallel

  kvareddy

  committed
  aae72ff

  Copy full SHA for aae72ff

• address review comment

  kvareddy

  committed
  214a9a0

  Copy full SHA for 214a9a0

• cleanup imports in transformer.py

  kvareddy

  committed
  28b6ffc

  Copy full SHA for 28b6ffc

• update jit warmup code to handle sequence parallelism

  kvareddy

  committed
  a4ef7c4

  Copy full SHA for a4ef7c4

• Merge branch 'main' into sequence_parallel

  kvareddy

  committed
  7f9a48b

  Copy full SHA for 7f9a48b

Commits on May 13, 2022

• address review comments

  kvareddy

  committed
  35a3fa8

  Copy full SHA for 35a3fa8

Commits on Apr 28, 2022

• avoid 'reverse checkpointed' term in argument description

  kvareddy

  committed
  0012901

  Copy full SHA for 0012901

• reverting pre allocation of optimizer states; it does not seem to help with fragmentation

  kvareddy

  committed
  5d2e13a

  Copy full SHA for 5d2e13a

• address review comments

  kvareddy

  committed
  cfd2e21

  Copy full SHA for cfd2e21

Pagination

Previous

Next