Commits · 4e6442903b511ae5dd398d8204cd9eebf0729c77 · Shashank Suhas / seminar-breakout

19 Oct, 2017 7 commits
- Add a function to apply general prefetch policies · 4e644290
  Yuxin Wu authored Oct 19, 2017
  
  4e644290
- Use `logger.get_logger_dir()` to access logger directory. · fff3f2d3
  Yuxin Wu authored Oct 19, 2017
  
  fff3f2d3
- fix bug of last commit · 1be49dc9
  Yuxin Wu authored Oct 19, 2017
  
  1be49dc9
- Builder with get_grads instead of input+get_cost · d2939cf8
  Yuxin Wu authored Oct 19, 2017
  
  d2939cf8
- Allow double-wrapping TowerFuncWrapper · 2c5b1bec
  Yuxin Wu authored Oct 18, 2017
  
  2c5b1bec
- deprecate TrainConfig(predict_tower). The device should be passed directly to InferenceRunner · d9817a56
  Yuxin Wu authored Oct 18, 2017
  
  d9817a56
- [Trainerv2] Let InferenceRunner run with new Trainer · da98e447
  Yuxin Wu authored Oct 18, 2017
  
  da98e447
18 Oct, 2017 15 commits
- a better TowerTensorHandle to access tensors (currently for predictor only) (#318) · e5ff50e7
  Yuxin Wu authored Oct 18, 2017
  
  e5ff50e7
- bugfix of last commit. · 9911b234
  Yuxin Wu authored Oct 18, 2017
  
  9911b234
- Fix imports; Add call_only_once decorator; · 260936ff
  Yuxin Wu authored Oct 18, 2017
  
  260936ff
- remove DataParallelFeedInput, as dataparallel inference now relies on QueueInput · 9fff46d5
  Yuxin Wu authored Oct 18, 2017
  
  9fff46d5
- [Trainerv2] fix interface for distributed trainer. · 12cd6e6c
  Yuxin Wu authored Oct 18, 2017
  
  12cd6e6c
- initial commit of new trainer interface (#318) · a673974c
  Yuxin Wu authored Oct 18, 2017
  
  a673974c
- Share documents between builder & trainer · 82187086
  Yuxin Wu authored Oct 18, 2017
  
  82187086
- bump versions · 5b8ed8be
  Yuxin Wu authored Oct 18, 2017
  
  5b8ed8be
- StagingInputWrapper takes a list of int · e6493857
  Yuxin Wu authored Oct 18, 2017
  
  e6493857
- Fix some use of loop attributes. These attributes should be available only from before_train · 4c5cdf9b
  Yuxin Wu authored Oct 18, 2017
  
  4c5cdf9b
- make _monitors private as well. · 4126a583
  Yuxin Wu authored Oct 17, 2017
  
  4126a583
- fix __doc__ assignment in Py3 · e10d43e7
  Yuxin Wu authored Oct 17, 2017
  
  e10d43e7
- update docs · c33a3ccb
  Yuxin Wu authored Oct 17, 2017
  
  c33a3ccb
- fix missing import · caf9ee8d
  Yuxin Wu authored Oct 17, 2017
  
  caf9ee8d
- move InputSource to a separate folder · 63c0f891
  Yuxin Wu authored Oct 17, 2017
  
  63c0f891
17 Oct, 2017 18 commits
- Use `TrainLoop` to manage the loop, and delegate properties. Hide `trainer.config` · 9e995a8d
  Yuxin Wu authored Oct 17, 2017
  
  9e995a8d
- fix missing chief setting in Distributed Trainer · 7efe4939
  Yuxin Wu authored Oct 17, 2017
  
  7efe4939
- extract common utilities out of train/ · 10cc1962
  Yuxin Wu authored Oct 17, 2017
  
  10cc1962
- fix deprecated usage in examples · a64d25cf
  Yuxin Wu authored Oct 17, 2017
  
  a64d25cf
- clean-ups · 12192d59
  Yuxin Wu authored Oct 17, 2017
  
  12192d59
- improve logging and fix wd computation in inference · f67a1aff
  Yuxin Wu authored Oct 17, 2017
  
  f67a1aff
- Maintain global_step in each job as a local_variable · 12846f57
  Yuxin Wu authored Oct 17, 2017
  
  12846f57
- clean-up ModelDesc · 3498c0b6
  Yuxin Wu authored Oct 17, 2017
  
  3498c0b6
- Docs about graph builder · 490142d7
  Yuxin Wu authored Oct 17, 2017
  
  490142d7
- add missing file · a2f60395
  Yuxin Wu authored Oct 17, 2017
  
  a2f60395
- fix distributed training, GAN examples. rename utilities · 6cb47609
  Yuxin Wu authored Oct 17, 2017
  
  6cb47609
- fix multigpu training · ce709fa3
  Yuxin Wu authored Oct 17, 2017
  
  ce709fa3
- [WIP] move away all graph building logic · 694e404b
  Yuxin Wu authored Oct 16, 2017
  
  694e404b
- remove feedfree · ad5cb725
  Yuxin Wu authored Oct 16, 2017
  
  ad5cb725
- graph builder for simple trainer · 82bf74c9
  Yuxin Wu authored Oct 16, 2017
  
  82bf74c9
- [FasterRCNN] restructure main function · 94eace54
  Yuxin Wu authored Oct 17, 2017
  
  94eace54
- [FasterRCNN] support 1,2,4 GPUs · 610bd283
  Yuxin Wu authored Oct 16, 2017
  
  610bd283
- RunOp accepts either Op or lambda · 3db6ccac
  Yuxin Wu authored Oct 17, 2017
  
  3db6ccac