Commits · 214a9a08697fcd6d38b36a0dd7304aea0b7e48c1 · OpenDAS / Megatron-LM

16 May, 2022 4 commits
- address review comment · 214a9a08
  Vijay Korthikanti authored May 16, 2022
  
  214a9a08
- cleanup imports in transformer.py · 28b6ffce
  Vijay Korthikanti authored May 16, 2022
  
  28b6ffce
- update jit warmup code to handle sequence parallelism · a4ef7c40
  Vijay Korthikanti authored May 16, 2022
  
  a4ef7c40
- Merge branch 'main' into sequence_parallel · 7f9a48ba
  Vijay Korthikanti authored May 16, 2022
  
  7f9a48ba
10 May, 2022 1 commit

Merge branch 'slym/jit_warmup' into 'main' · d2394294

Jared Casper authored May 10, 2022

jit functions warmpups to (1) match fprop and recompute results and (2) remove grad_enable

See merge request ADLR/megatron-lm!404

d2394294

09 May, 2022 1 commit
- (1)support pyt20.12 compatibility, (2) arg naming update · f602ac56
  Sangkug Lym authored Apr 11, 2022
```
change dummy_handler to nullcontext
```
  f602ac56
28 Apr, 2022 3 commits
- avoid 'reverse checkpointed' term in argument description · 00129014
  Vijay Korthikanti authored Apr 28, 2022
  
  00129014
- reverting pre allocation of optimizer states; it does not seem to help with fragmentation · 5d2e13a6
  Vijay Korthikanti authored Apr 28, 2022
  
  5d2e13a6
- address review comments · cfd2e216
  Vijay Korthikanti authored Apr 28, 2022
  
  cfd2e216
25 Apr, 2022 1 commit
- make interleaving work with optimizations · 13b3dca6
  Vijay Korthikanti authored Apr 25, 2022
  
  13b3dca6
30 Mar, 2022 1 commit
- sequence parallelism for embedding dropout and last linear layer + memory optimizations · eec218d8
  Vijay Korthikanti authored Mar 30, 2022
  
  eec218d8
25 Mar, 2022 1 commit
- jit functions warmpups to (1) match fprop and recompute results and (2) remove grad_enable · c7fef593
  Sangkug Lym authored Mar 24, 2022
  
  c7fef593
15 Mar, 2022 3 commits
- Merge branch 'lmcafee/free-pipe-output-bug' into 'main' · 15f6bb1b
  Jared Casper authored Mar 15, 2022
```
Bugfix: Pipeline output deallocation w/ persistent layer norm

See merge request ADLR/megatron-lm!402
```
  15f6bb1b
- fixed. · 65e6bc32
  Lawrence McAfee authored Mar 15, 2022
  
  65e6bc32
- debugging. · 4c598f9d
  Lawrence McAfee authored Mar 15, 2022
  
  4c598f9d
11 Mar, 2022 1 commit
- checkpoint support for core attention · 53718d4c
  Vijay Korthikanti authored Mar 11, 2022
  
  53718d4c
09 Mar, 2022 1 commit
- bug fixes · dd96d402
  Vijay Korthikanti authored Mar 08, 2022
  
  dd96d402
07 Mar, 2022 3 commits
- fixes to main merge · 269f28f7
  Vijay Korthikanti authored Mar 07, 2022
  
  269f28f7
- Merge branch 'main' into allreduce-split · 6fdbf26b
  Vijay Korthikanti authored Mar 07, 2022
  
  6fdbf26b
- refactor to help merge with main · 0d77c0e9
  Vijay Korthikanti authored Mar 07, 2022
  
  0d77c0e9
03 Mar, 2022 2 commits
- column parallel linear with sequence parallelism · 02bb1f5c
  Vijay Korthikanti authored Mar 02, 2022
  
  02bb1f5c
- get sequence parallelism to work with pipeline parallelism · 6658158b
  Vijay Korthikanti authored Mar 02, 2022
  
  6658158b
02 Mar, 2022 1 commit
- layernorm grad sync + name chnages · c0f10643
  Vijay Korthikanti authored Mar 02, 2022
  
  c0f10643
24 Feb, 2022 2 commits
- Merge branch 'switch' into 'main' · 53f3efc4
  Lawrence McAfee authored Feb 24, 2022
```
Switch Transformer

See merge request ADLR/megatron-lm!373
```
  53f3efc4
- Fixing merge conflict · 458d7785
  rprenger authored Feb 23, 2022
  
  458d7785
23 Feb, 2022 1 commit
- Merge branch 'slym/grad_accum_fusion' into 'main' · 9c5a830f
  Jared Casper authored Feb 22, 2022
```
Gradient accumulation fusion

See merge request ADLR/megatron-lm!394
```
  9c5a830f
19 Feb, 2022 1 commit
- tensor model parallelism memory optmization · 5d4689c4
  Vijay Korthikanti authored Feb 18, 2022
  
  5d4689c4
18 Feb, 2022 1 commit
- support fp32 training and fix embedding update · b5726555
  Sangkug Lym authored Feb 18, 2022
  
  b5726555
17 Feb, 2022 6 commits
- Merge branch 'all_gather_base' into 'main' · 0ed2f6ac
  Jared Casper authored Feb 17, 2022
```
changed all_gather to _all_gather_base in distributed checkpointing

See merge request ADLR/megatron-lm!395
```
  0ed2f6ac
- Don't need an example for Switch networks until it's more polished · d7bf1ab5
  rprenger authored Feb 17, 2022
  
  d7bf1ab5
- addressed Jared and Patrick comments. · 90ce932d
  mshoeybi authored Feb 17, 2022
  
  90ce932d
- changed all_gather to _all_gather_base in distributed checkpointing · 37181ef4
  mshoeybi authored Feb 17, 2022
  
  37181ef4
- removed slurm commands; added 'num-experts' · 3237cc33
  Lawrence McAfee authored Feb 17, 2022
  
  3237cc33
- moved run.sh -> examples/ · c8e4cdae
  Lawrence McAfee authored Feb 17, 2022
  
  c8e4cdae
16 Feb, 2022 1 commit

gradient accumulation fusion · 83b1e42f

Sangkug Lym authored Feb 13, 2022

remove redundant linear layer class definition

add fuse_gradient_accumulation attribute to weights for simple targetting

reflect feedback and clean up the codes

arg change

83b1e42f

15 Feb, 2022 5 commits
- adress review comments · 488f8c02
  Vijay Korthikanti authored Feb 11, 2022
  
  488f8c02
- more minor fixes · 1cd3650d
  Vijay Korthikanti authored Feb 01, 2022
  
  1cd3650d
- minor fixes · 6f3bf9c0
  Vijay Korthikanti authored Feb 01, 2022
  
  6f3bf9c0
- vision third phase merge: pretraining methods + mit,swin backbones · 48c2a144
  Vijay Korthikanti authored Jan 31, 2022
  
  48c2a144
- Merge branch 'vision-merge' into 'main' · 8c8063eb
  Jared Casper authored Feb 14, 2022
```
vision third phase merge: pretraining methods + mit,swin backbones

See merge request ADLR/megatron-lm!384
```
  8c8063eb