Commits · e9d46fa0fe7cf69fb00d7eab7028407b57be378f · OpenDAS / Uni-Core

29 Mar, 2023 1 commit
- fix loading model logs (#24) · e9d46fa0
  RUCer authored Mar 29, 2023
```
* fix log

* fix log
```
  e9d46fa0
09 Mar, 2023 1 commit
- pass arg to switch ckp on/off (#22) · a44b9c68
  Ziyao Li authored Mar 09, 2023
  
  a44b9c68
05 Mar, 2023 2 commits
- fix numpy seed key ignorance (#20) · ee13d84f
  Ziyao Li authored Mar 05, 2023
  
  ee13d84f
- support layernorm with dim 192 · 3aa106dc
  Guolin Ke authored Mar 05, 2023
  
  3aa106dc
23 Feb, 2023 1 commit
- filter params in ema (#19) · f2b0b488
  Jinhua Zhu authored Feb 23, 2023
  
  f2b0b488
12 Jan, 2023 2 commits

fix validate_with_ema option · 4e4dc4b8
Guolin Ke authored Jan 12, 2023

4e4dc4b8

Jinhua Zhu authored Jan 12, 2023



* valid with ema

* save ema model

* update ema

* rm model_fp32

* rm ordered names

* load ema

* emapy

* Update unicore/trainer.py
Co-authored-by: Guolin Ke <guolin.ke@outlook.com>

d01a62ef

26 Sep, 2022 1 commit
- fix ckp loading logging · f24a5f70
  Guolin Ke authored Sep 26, 2022
  
  f24a5f70
08 Sep, 2022 1 commit
- fix logging in load checkpoint · 486f4f3d
  Guolin Ke authored Sep 08, 2022
  
  486f4f3d
05 Sep, 2022 2 commits
- refine rngstate in trainer · 4c06a4fd
  Guolin Ke authored Sep 05, 2022
  
  4c06a4fd
- bug fix for rngstate with update_freq > 1. · 70dcdf17
  Guolin Ke authored Sep 05, 2022
  
  70dcdf17
27 Aug, 2022 1 commit
- rollback to torch when check bias/mask failed in softmax · 0a79672a
  Guolin Ke authored Aug 27, 2022
  
  0a79672a
26 Aug, 2022 2 commits
- fix a possible type bug in layernorm · f4ce5889
  Guolin Ke authored Aug 26, 2022
  
  f4ce5889
- remove pandas · f913d977
  Guolin Ke authored Aug 26, 2022
  
  f913d977
24 Aug, 2022 2 commits
- typo · 83602797
  Guolin Ke authored Aug 24, 2022
  
  83602797
- more layer_norm kernels · cf503760
  Guolin Ke authored Aug 24, 2022
  
  cf503760
14 Aug, 2022 2 commits
- refine docker build (#9) · 5551a790
  Guolin Ke authored Aug 14, 2022
```
* Update docker.yml

* py version >= 3.7

* rename docker image

* update docker file
```
  5551a790
- support old GPUS like p100 (#7) · ccf3bea4
  Guolin Ke authored Aug 14, 2022
  
  ccf3bea4
12 Aug, 2022 1 commit
- support softmax in non-inplace cases · af4f9088
  Guolin Ke authored Aug 12, 2022
  
  af4f9088
11 Aug, 2022 3 commits
- fix bug in return future mask · 49c9895b
  Guolin Ke authored Aug 11, 2022
  
  49c9895b
- fix a bug for return attn_weights · 689e0b24
  Guolin Ke authored Aug 11, 2022
  
  689e0b24
- fix bug in flatten parameters · a92c6297
  Guolin Ke authored Aug 11, 2022
  
  a92c6297
04 Aug, 2022 1 commit
- support softmax for large columns (#6) · 6c18b2ed
  Guolin Ke authored Aug 04, 2022
```
* support softmax for large columns

* more tests
```
  6c18b2ed
30 Jul, 2022 1 commit
- support gradient checkpointing · 31fe887e
  Guolin Ke authored Jul 30, 2022
  
  31fe887e
28 Jul, 2022 1 commit
- flatten parameters (#4) · 89ac1b7b
  Guolin Ke authored Jul 28, 2022
```
flatten memories
```
  89ac1b7b
27 Jul, 2022 1 commit
- fix softmax cpu · 1e8b6e33
  Guolin Ke authored Jul 27, 2022
  
  1e8b6e33
24 Jul, 2022 2 commits
- warning about missing keys in loading model · 21cb6b39
  Guolin Ke authored Jul 24, 2022
  
  21cb6b39
- support softmax with bias and mask (#2) · 8da5eaaf
  Guolin Ke authored Jul 24, 2022
```
* try

* fix bugs

* code clean

* support mask in softmax

* code clean

* check for shapes
```
  8da5eaaf
22 Jul, 2022 1 commit
- use legacy pytorch fuser · 0da9683c
  Guolin Ke authored Jul 22, 2022
  
  0da9683c
19 Jul, 2022 1 commit
- remaning no_ to num_ · a6ed694a
  Guolin Ke authored Jul 19, 2022
  
  a6ed694a
13 Jul, 2022 1 commit
- save memory for softmax · 3f498d32
  Guolin Ke authored Jul 13, 2022
  
  3f498d32
10 Jul, 2022 2 commits
- Update fused_adam.py · 979f972a
  Guolin Ke authored Jul 10, 2022
  
  979f972a
- first commit · 8a915496
  Guolin Ke authored Jul 10, 2022
  
  8a915496