Commits · 7db9c1e55c9739adf2e52dbc05f7bcb3e5ee647e · OpenDAS / Megatron-LM

25 Jan, 2021 2 commits
- Use set_tensor_model_parallel_attributes in bert_model as well. · c4c68dce
  Jared Casper authored Jan 25, 2021
  
  c4c68dce
- Address comments, fix argument bug. · bdd47d64
  Jared Casper authored Jan 25, 2021
  
  bdd47d64
22 Jan, 2021 1 commit
- Fixing merge_mp_partitions · 78066ab0
  Jared Casper authored Jan 20, 2021
  
  78066ab0
04 Jan, 2021 1 commit
- Use batched send and recv instead of torch.distributed.ring_exchange() · d899988e
  Deepak Narayanan authored Jan 04, 2021
  
  d899988e
25 Dec, 2020 3 commits
- grads is removed from mpu · dfd8ed47
  mohammad authored Dec 25, 2020
  
  dfd8ed47
- clip grad fixed and moved to optimizer · 46879674
  mohammad authored Dec 25, 2020
  
  46879674
- fp32 is also working · 2eaa3ccc
  mohammad authored Dec 24, 2020
  
  2eaa3ccc
22 Dec, 2020 1 commit
- Add the option for fp32 residual connection (fp32 residual connection... · 62632d39
  mohammad authored Dec 21, 2020
```
Add the option for fp32 residual connection (fp32 residual connection machinery still needs to be added)
```
  62632d39
19 Dec, 2020 4 commits
- Add comment describing _PIPELINE_GLOBAL_RANKS · 51315905
  Jared Casper authored Dec 10, 2020
  
  51315905
- Initial implementation of pipelined text generation · 5c45db4a
  Jared Casper authored Dec 09, 2020
  
  5c45db4a
- Add constant num micro-batches calculator · feecd5d9
  mohammad authored Dec 07, 2020
  
  feecd5d9
- Rename --batch-size to --micro-batch-size and drop in-minibatch from... · 9019bbf4
  mohammad authored Dec 06, 2020
```
Rename --batch-size to --micro-batch-size and drop in-minibatch from --num-micro-batches-in-minibatch
```
  9019bbf4
18 Nov, 2020 1 commit
- Replace deprecated torch.norm with torch.linalg.norm. · 17035d6c
  Jared Casper authored Nov 17, 2020
  
  17035d6c
12 Nov, 2020 3 commits
- Back compatibility of checkpoints: don't rename model_parallel_rng_tracker · e805f0bd
  Deepak Narayanan authored Oct 22, 2020
  
  e805f0bd
- Intra-layer MP -> Tensor MP, Inter-layer MP -> Pipeline MP · 52a5f2f2
  Deepak Narayanan authored Oct 20, 2020
  
  52a5f2f2
- Pipeline parallelism implementation with periodic full-pipeline syncs · 7abd3e90
  Deepak Narayanan authored Aug 29, 2020
```
Also includes following changes for inter-layer model-parallel implementation:
- Refactoring of model implementations
- Training loop changes to support inter-layer communication using `ring_exchange`
- New groups for inter-layer communication
- Checkpoint changes
- Command line arguments
```
  7abd3e90
02 Oct, 2020 1 commit
- Fixed uninit communication gruop · 057a1fab
  Mostofa Patwary authored Oct 01, 2020
  
  057a1fab
11 Sep, 2020 1 commit
- Various speed optimizations. · 51a2e6b0
  Vijay Korthikanti authored Sep 11, 2020
  
  51a2e6b0
03 Sep, 2020 1 commit
- debuged the cpu init for mp=1 · 04cd82bd
  mohammad authored Sep 03, 2020
  
  04cd82bd
02 Sep, 2020 2 commits
- addressed Jareds comments · 5d29769c
  mohammad authored Sep 02, 2020
  
  5d29769c
- added splitting checkpointed activations · d6c4248b
  mohammad authored Sep 02, 2020
  
  d6c4248b
27 Aug, 2020 1 commit
- Addressing code review comments · 1eafa861
  Boris Fomitchev authored Aug 26, 2020
```
Signed-off-by: Boris Fomitchev <bfomitchev@nvidia.com>
```
  1eafa861
26 Aug, 2020 1 commit
- Making --lazy_mpu_init act as _USE_CPU_INITIALIZATION · e10760ea
  Boris Fomitchev authored Aug 26, 2020
  
  e10760ea
07 Aug, 2020 2 commits
- Fix typo · 9edfdaec
  Nako Sung authored Aug 08, 2020
```
redcue -> reduce
```
  9edfdaec
- Fix typo · f63a8a12
  Nako Sung authored Aug 08, 2020
```
Tiny typo fix : sedd -> seed
```
  f63a8a12
06 Aug, 2020 1 commit
- Implementing lazy parallel initialization · 5c04ceb3
  Boris Fomitchev authored Aug 05, 2020
```
Signed-off-by: Boris Fomitchev <bfomitchev@nvidia.com>
```
  5c04ceb3
03 Aug, 2020 1 commit
- added gpu initialization and option to avoid master values · 0403b808
  mohammad authored Aug 03, 2020
  
  0403b808
31 Jul, 2020 1 commit
- Code review comments - changing parallel test condition · 35bea728
  Boris Fomitchev authored Jul 30, 2020
```
Signed-off-by: Boris Fomitchev <bfomitchev@nvidia.com>
```
  35bea728
30 Jul, 2020 1 commit
- Changes for NeMo/lightning compatibility · 417c7f6a
  Boris Fomitchev authored Jul 30, 2020
```
Signed-off-by: Boris Fomitchev <bfomitchev@nvidia.com>
```
  417c7f6a
26 Jul, 2020 2 commits
- Add additional assertion on Indexer to test correctness, and limit verbosity in other classes · eaa5d877
  Neel Kant authored Jul 25, 2020
  
  eaa5d877
- Complete fix for APEX absence in NeMo · aa0ee72e
  Boris Fomitchev authored Jul 25, 2020
  
  aa0ee72e
24 Jul, 2020 1 commit
- ONNX export fix, including one for gelu() · 158a99c3
  Boris Fomitchev authored Jul 23, 2020
```
Signed-off-by: Boris Fomitchev <bfomitchev@nvidia.com>
```
  158a99c3
04 Jul, 2020 1 commit
- small fix for CheckpointFunction's backward() method when sone args may be NoneType · 4ee0537a
  Devendra Singh Sachan authored Jul 04, 2020
  
  4ee0537a
05 Jun, 2020 1 commit
- Prune changes to only be related to ICT · 32bb4edc
  Neel Kant authored Jun 05, 2020
  
  32bb4edc
26 May, 2020 1 commit
- Corrected realm example building, misc improvements for async concurrency · 2fd4ea6c
  Neel Kant authored May 25, 2020
  
  2fd4ea6c
21 May, 2020 1 commit
- Change sync variable to gloo backend · 05ea0cca
  Neel Kant authored May 21, 2020
  
  05ea0cca
20 May, 2020 1 commit
- Async works for total 8 GPU, indexer debug mode · a670b6c9
  Neel Kant authored May 19, 2020
  
  a670b6c9
19 May, 2020 2 commits
- Reorganize indexer. Things run up to saving model checkpoint and repeating · d4b00be0
  Neel Kant authored May 19, 2020
  
  d4b00be0
- Indexer_async works in theory · e338e311
  Neel Kant authored May 18, 2020
  
  e338e311
21 Apr, 2020 1 commit
- removed optimized l2 grad clipping · 0b6bdca6
  Mohammad authored Apr 21, 2020
  
  0b6bdca6