Commits

Breadcrumbs

History for

onmaster

User selector

All users

All time

Commit History

Commits on May 2, 2019

• [src] Keep nnet output in the [-30,30] range required by chain denominator (#3024)

  Show description for b1569db
  danpovey

  authored
  b1569db

  Copy full SHA for b1569db

Commits on Oct 7, 2018

• [src] Some minor cleanup/fixes regarding CUDA memory allocation; other small fixes. (#2768)

  danpovey

  authored
  e54462e

  Copy full SHA for e54462e

Commits on Oct 4, 2018

• [src,scripts] Batched nnet3 computation (#2726)

  Show description for c78875a
  danpovey

  authored
  c78875a

  Copy full SHA for c78875a

Commits on Sep 19, 2018

• [scripts,src] Fix potential issue in scripts; minor fixes. (#2724)

  Show description for 69cd717
  danpovey

  authored
  69cd717

  Copy full SHA for 69cd717

Commits on Aug 8, 2018

• [src] Finish removing old context-fst code and shift calling code to use the new mechanism.

  danpovey

  committed
  a3a547c

  Copy full SHA for a3a547c

Commits on Jun 19, 2018

• [src] Minor optimizations in "e2e" numerator code (#2508)

  

  hhadian

  authored and

  danpovey

  committed
  775c770

  Copy full SHA for 775c770

Commits on Jun 5, 2018

• [src,scripts] Add steps/nnet3/chain/get_phone_post.sh, to get phone posteriors from chain system (#2471)

  danpovey

  authored
  ce2eb39

  Copy full SHA for ce2eb39

Commits on May 7, 2018

• [src] make e2e/"unconstrained" numerator computation faster (#2392)

  

  jtrmal

  authored and

  danpovey

  committed
  954e69a

  Copy full SHA for 954e69a

Commits on May 3, 2018

• [src] Remove unnecessary file chain/context-dep-topology.h (#2401)

  

  huangruizhe

  authored and

  danpovey

  committed
  20b2a45

  Copy full SHA for 20b2a45

Commits on Apr 29, 2018

• [src] Remove un-needed code. Thanks: [email protected]. (#2393)

  

  huangruizhe

  authored and

  danpovey

  committed
  bf40d61

  Copy full SHA for bf40d61

Commits on Apr 25, 2018

• [src] Add a cast to prevent errors from some compilers (#2384)

  

  danijel3

  authored and

  danpovey

  committed
  d25b3ef

  Copy full SHA for d25b3ef

Commits on Apr 24, 2018

• [src,scripts,egs] Support for unconstrained-egs for chain training + example recipes (#2383)

  Show description for f0333bb
  

  hhadian

  authored and

  danpovey

  committed
  f0333bb

  Copy full SHA for f0333bb

Commits on Mar 28, 2018

• [src,scripts,egs] Semi-supervised training on Fisher English (#2140)

  

  vimalmanohar

  authored and

  danpovey

  committed
  191b39a

  Copy full SHA for 191b39a

• [src] Apply limits prior to chain denominator computation, avoid failures. (#2308)

  danpovey

  authored
  8af60bb

  Copy full SHA for 8af60bb

Commits on Mar 23, 2018

• [src] Speedup and code simplification for chain supervision merging (etc.) (#2302)

  

  chenzhehuai

  authored and

  danpovey

  committed
  6dbe790

  Copy full SHA for 6dbe790

Commits on Mar 12, 2018

• [src] bug-fixes for end2end chain code (#2270)

  

  hhadian

  authored and

  danpovey

  committed
  19dc26f

  Copy full SHA for 19dc26f

Commits on Mar 6, 2018

• [src,scripts,egs] End-to-end chain training (#2072)

  Show description for e74b918
  

  hhadian

  authored and

  danpovey

  committed
  e74b918

  Copy full SHA for e74b918

Commits on Mar 1, 2018

• [src,scripts,egs] Add form of dropout that shares the mask across frames. (#2244)

  danpovey

  authored
  9a4ba5e

  Copy full SHA for 9a4ba5e

Commits on Jan 30, 2018

• [src] Various bug-fixes relating to recent nnet3/chain changes.

  danpovey

  committed
  cf74573

  Copy full SHA for cf74573

Commits on Jan 29, 2018

• [src] Change for memory efficiency to chain training; small cosmetic fixes.

  danpovey

  committed
  8638ec9

  Copy full SHA for 8638ec9

• [src] Some small cosmetic changes

  danpovey

  committed
  b9fc151

  Copy full SHA for b9fc151

Commits on May 26, 2017

• [src] Use STL thread support library instead of pthread. (#1350)

  

  dogancan

  authored and

  danpovey

  committed
  6cc8e3a

  Copy full SHA for 6cc8e3a

Commits on May 13, 2017

• [src] Change to GPU synchronization, for speed (disables GPU stats by default) (#1617)

  danpovey

  authored
  b1e8601

  Copy full SHA for b1e8601

Commits on Apr 2, 2017

• [build,src,doc] Modify get_version.sh to deal better with whitespace (avoid space in version); minor fixes (#1526)

  danpovey

  authored
  60f9327

  Copy full SHA for 60f9327

Commits on Feb 17, 2017

• [src] Make chain-supervision-test.cc faster when no GPU

  danpovey

  committed
  deda2bd

  Copy full SHA for deda2bd

Commits on Jan 23, 2017

• [src] speed up test chain/chain-supervision-test.cc

  danpovey

  committed
  44fa7ee

  Copy full SHA for 44fa7ee

Commits on Jan 18, 2017

• Upgrade codebase to support OpenFst-1.6.0.

  dogancan

  committed
  e2ecec4

  Copy full SHA for e2ecec4

Commits on Dec 15, 2016

• Remove no-longer-used option --cut-zero-frames from chain supervision-creation code

  danpovey

  committed
  60f79dc

  Copy full SHA for 60f79dc

Commits on Nov 30, 2016

• Fixes to CUDA problems introduced by recent commits: fix normalize code crash; compile problem on nvcc 8.0; fix thread-sync errors. (#1228)

  Show description for 87465f5
  danpovey

  authored
  87465f5

  Copy full SHA for 87465f5

Commits on Nov 26, 2016

• use native double atomicAdd for CC6.0 and devices; fix bugs;

  kangshiyin

  committed
  0cbc041

  Copy full SHA for 0cbc041

Commits on Nov 21, 2016

• update comments for new transition model

  naxingyu

  committed
  168957c

  Copy full SHA for 168957c

• modify chain code for new transition model

  naxingyu

  committed
  33957ab

  Copy full SHA for 33957ab

Commits on Oct 11, 2016

• Unk model (#1058)

  Show description for a8de21f
  danpovey

  authored
  a8de21f

  Copy full SHA for a8de21f

Commits on Sep 14, 2016

• Replace implementation of atomic addition.

  Show description for 6f20b39
  galv

  committed
  6f20b39

  Copy full SHA for 6f20b39

Commits on Sep 5, 2016

• Removing little-used feature: time-reversed, and forward-backward, decoding.

  danpovey

  committed
  5dfa20a

  Copy full SHA for 5dfa20a