github
/
ColossalAI
mirror of https://github.com/hpcaitech/ColossalAI
1
0
Fork 0
Code Issues Projects Releases Wiki Activity
1,918 Commits
31 Branches
47 Tags
229 MiB
Commit Graph

1231 Commits (4ae02c4b1c9a9ac710b11a08cc166aaa90944d5b)

Author SHA1 Message Date
oahzxl 6ba8364881
[autochunk] support diffusion for autochunk (#2621) 
* add alphafold benchmark

* renae alphafold test

* rename tests

* rename diffuser

* renme

* rename

* update transformer

* update benchmark

* update benchmark

* update bench memory

* update transformer benchmark

* rename

* support diffuser

* support unet metainfo prop

* fix bug and simplify code

* update linear and support some op

* optimize max region search, support conv

* update unet test

* support some op

* support groupnorm and interpolate

* update flow search

* add fix dim in node flow

* fix utils

* rename

* support diffusion

* update diffuser

* update chunk search

* optimize imports

* import

* finish autochunk
 2023-02-07 16:32:45 +08:00
Frank Lee 8518263b80
[test] fixed the triton version for testing (#2608) 2023-02-07 13:49:38 +08:00
HELSON 552183bb74
[polish] polish ColoTensor and its submodules (#2537) 2023-02-03 11:44:10 +08:00
Frank Lee dd14783f75
[kernel] fixed repeated loading of kernels (#2549) 
* [kernel] fixed repeated loading of kernels

* polish code

* polish code
 2023-02-03 09:47:13 +08:00
ver217 5b1854309a
[hotfix] fix zero ddp warmup check (#2545) 2023-02-02 16:42:38 +08:00
oahzxl fa3d66feb9
support unet metainfo prop (#2544) 2023-02-02 16:19:26 +08:00
oahzxl 05671fcb42
[autochunk] support multi outputs chunk search (#2538) 
Support multi outputs chunk search. Previously we only support single output chunk search. It is more flexible and improve performance by a large margin. For transformer, we reduce memory by 40% than previous search strategy.

1. rewrite search strategy to support multi outputs chunk search
2. fix many, many bugs
3. update tests
 2023-02-01 13:18:51 +08:00
oahzxl 63199c6687
[autochunk] support transformer (#2526) 2023-01-31 16:00:06 +08:00
HELSON a4ed9125ac
[hotfix] fix lightning error (#2529) 2023-01-31 10:40:39 +08:00
HELSON 66dfcf5281
[gemini] update the gpt example (#2527) 2023-01-30 17:58:05 +08:00
HELSON b528eea0f0
[zero] add zero wrappers (#2523) 
* [zero] add zero wrappers

* change names

* add wrapper functions to init
 2023-01-29 17:52:58 +08:00
Super Daniel c198c7c0b0
[hotfix] meta tensor default device. (#2510) 2023-01-29 16:28:10 +08:00
HELSON 077a5cdde4
[zero] fix gradient clipping in hybrid parallelism (#2521) 
* [zero] fix gradient clipping in hybrid parallelism

* [testing] change model name to avoid pytest warning

* [hotfix] fix unit testing
 2023-01-29 15:09:57 +08:00
YuliangLiu0306 aa0f6686f9
[autoparallel] accelerate gpt2 training (#2495) 2023-01-29 11:13:15 +08:00
HELSON 707b11d4a0
[gemini] update ddp strict mode (#2518) 
* [zero] add strict ddp mode for chunk init

* [gemini] update gpt example
 2023-01-28 14:35:25 +08:00
HELSON 2d1a7dfe5f
[zero] add strict ddp mode (#2508) 
* [zero] add strict ddp mode

* [polish] add comments for strict ddp mode

* [zero] fix test error
 2023-01-20 14:04:38 +08:00
oahzxl c04f183237
[autochunk] support parsing blocks (#2506) 2023-01-20 11:18:17 +08:00
Super Daniel 35c0c0006e
[utils] lazy init. (#2148) 
* [utils] lazy init.

* [utils] remove description.

* [utils] complete.

* [utils] finalize.

* [utils] fix names.
 2023-01-20 10:49:00 +08:00
oahzxl 72341e65f4
[auto-chunk] support extramsa (#3) (#2504) 2023-01-20 10:13:03 +08:00
Ziyue Jiang 0f02b8c6e6
add avg partition (#2483) 
Co-authored-by: Ziyue Jiang <ziyue.jiang@gmail.com>
 2023-01-19 13:54:50 +08:00
アマデウス 99d9713b02 Revert "Update parallel_context.py (#2408)" 
This reverts commit 7d5640b9db.
 2023-01-19 12:27:48 +08:00
oahzxl ecccc91f21
[autochunk] support autochunk on evoformer (#2497) 2023-01-19 11:41:00 +08:00
oahzxl 5db3a5bf42
[fx] allow control of ckpt_codegen init (#2498) 
* [fx] allow control of ckpt_codegen init

Currently in ColoGraphModule, ActivationCheckpointCodeGen will be set automatically in __init__. But other codegen can't be set if so. 
So I add an arg to control whether to set ActivationCheckpointCodeGen in __init__.

* code style
 2023-01-18 17:02:46 +08:00
HELSON d565a24849
[zero] add unit testings for hybrid parallelism (#2486) 2023-01-18 10:36:10 +08:00
oahzxl 4953b4ace1
[autochunk] support evoformer tracer (#2485) 
support full evoformer tracer, which is a main module of alphafold. previously we just support a simplifed version of it.
1. support some evoformer's op in fx
2. support evoformer test
3. add repos for test code
 2023-01-16 19:25:05 +08:00
YuliangLiu0306 67e1912b59
[autoparallel] support origin activation ckpt on autoprallel system (#2468) 2023-01-16 16:25:13 +08:00
Ziyue Jiang fef5c949c3
polish pp middleware (#2476) 
Co-authored-by: Ziyue Jiang <ziyue.jiang@gmail.com>
 2023-01-13 16:56:01 +08:00
HELSON a5dc4253c6
[zero] polish low level optimizer (#2473) 2023-01-13 14:56:17 +08:00
Frank Lee 8b7495dd54
[example] integrate seq-parallel tutorial with CI (#2463) 2023-01-13 14:40:05 +08:00
Jiarui Fang 867c8c2d3a
[zero] low level optim supports ProcessGroup (#2464) 2023-01-13 10:05:58 +08:00
Frank Lee 14d9299360
[cli] fixed hostname mismatch error (#2465) 2023-01-12 14:52:09 +08:00
Haofan Wang 9358262992
Fix False warning in initialize.py (#2456) 
* Update initialize.py

* pre-commit run check
 2023-01-12 13:49:01 +08:00
YuliangLiu0306 8221fd7485
[autoparallel] update binary elementwise handler (#2451) 
* [autoparallel] update binary elementwise handler

* polish
 2023-01-12 09:35:10 +08:00
HELSON 2bfeb24308
[zero] add warning for ignored parameters (#2446) 2023-01-11 15:30:09 +08:00
Frank Lee 39163417a1
[example] updated the hybrid parallel tutorial (#2444) 
* [example] updated the hybrid parallel tutorial

* polish code
 2023-01-11 15:17:17 +08:00
HELSON 5521af7877
[zero] fix state_dict and load_state_dict for ddp ignored parameters (#2443) 
* [ddp] add is_ddp_ignored

[ddp] rename to is_ddp_ignored

* [zero] fix state_dict and load_state_dict

* fix bugs

* [zero] update unit test for ZeroDDP
 2023-01-11 14:55:41 +08:00
YuliangLiu0306 2731531bc2
[autoparallel] integrate device mesh initialization into autoparallelize (#2393) 
* [autoparallel] integrate device mesh initialization into autoparallelize

* add megatron solution

* update gpt autoparallel examples with latest api

* adapt beta value to fit the current computation cost
 2023-01-11 14:03:49 +08:00
Frank Lee c72c827e95
[cli] provided more details if colossalai run fail (#2442) 2023-01-11 13:56:42 +08:00
Super Daniel c41e59e5ad
[fx] allow native ckpt trace and codegen. (#2438) 2023-01-11 13:49:59 +08:00
YuliangLiu0306 41429b9b28
[autoparallel] add shard option (#2423) 2023-01-11 13:40:33 +08:00
HELSON 7829aa094e
[ddp] add is_ddp_ignored (#2434) 
[ddp] rename to is_ddp_ignored
 2023-01-11 12:22:45 +08:00
HELSON bb4e9a311a
[zero] add inference mode and its unit test (#2418) 2023-01-11 10:07:37 +08:00
Jiarui Fang 93f62dd152
[autochunk] add autochunk feature 2023-01-10 16:04:42 +08:00
HELSON dddacd2d2c
[hotfix] add norm clearing for the overflow step (#2416) 2023-01-10 15:43:06 +08:00
oahzxl 7ab2db206f adapt new fx 2023-01-10 11:56:00 +08:00
oahzxl e532679c95 Merge branch 'main' of https://github.com/oahzxl/ColossalAI into chunk 2023-01-10 11:29:01 +08:00
Haofan Wang 7d5640b9db
Update parallel_context.py (#2408) 2023-01-10 11:27:23 +08:00
oahzxl fd818cf144 change imports 2023-01-10 11:10:45 +08:00
oahzxl a591d45b29 add available 2023-01-10 10:56:39 +08:00
oahzxl 615e7e68d9 update doc 2023-01-10 10:44:07 +08:00