yh_cc
|
b0db3a5998
|
Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0
|
2022-04-15 19:31:58 +08:00 |
|
yh_cc
|
687db6d86a
|
1.torch在保存和load的时候会考虑GradScaler的保存问题; 2.新增Torch的GradientClip和Warmpup
|
2022-04-15 19:31:46 +08:00 |
|
x54-729
|
9e97155312
|
PaddleSingleDriver的save load函数测试
|
2022-04-15 10:44:49 +00:00 |
|
x54-729
|
be24572b11
|
修改paddle.distributed的import名
|
2022-04-15 10:44:25 +00:00 |
|
x54-729
|
16cec4bd99
|
删除不必要的测试文件
|
2022-04-15 09:07:35 +00:00 |
|
x54-729
|
cf19062fb2
|
set_dist_repro_dataloader测试例的完善
|
2022-04-15 09:06:22 +00:00 |
|
x54-729
|
288eb36afb
|
断点重训 save时的逻辑修正
|
2022-04-15 09:05:44 +00:00 |
|
x54-729
|
cca265f99c
|
Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0
|
2022-04-15 08:10:19 +00:00 |
|
x54-729
|
d0f26c7c34
|
将validate_step替换为evaluate_step
|
2022-04-15 08:10:01 +00:00 |
|
x54-729
|
8fc4fd19ff
|
small
|
2022-04-15 08:05:39 +00:00 |
|
yh_cc
|
7c70874b4a
|
删除core.sampelrs.sampler.py;增加torch的clipgradient和warmupcallback
|
2022-04-15 16:04:43 +08:00 |
|
x54-729
|
9d50e99bfb
|
修改evaluate_dataloader的报错信息
|
2022-04-15 06:57:31 +00:00 |
|
MorningForest
|
3ea74b52d2
|
Merge remote-tracking branch 'refs/remotes/origin/dev0.8.0' into dev0.8.0
|
2022-04-15 14:00:39 +08:00 |
|
yh_cc
|
f27d53261c
|
修改部分测试用例中validate_dataloader为evaluate_dataloader
|
2022-04-15 12:30:20 +08:00 |
|
YWMditto
|
02e080d239
|
Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0
|
2022-04-15 00:18:27 +08:00 |
|
YWMditto
|
3bbf27283d
|
little change
|
2022-04-15 00:17:50 +08:00 |
|
x54-729
|
f6f489dc90
|
多卡 set_dist_repro_dataloader 的测试例
|
2022-04-14 16:17:20 +00:00 |
|
YWMditto
|
a61f3975d7
|
little change
|
2022-04-15 00:17:01 +08:00 |
|
x54-729
|
98644d2d0b
|
fix conflict
|
2022-04-14 16:05:53 +00:00 |
|
x54-729
|
b97962b8dd
|
简化paddle trainer的单卡测试例
|
2022-04-14 16:04:56 +00:00 |
|
x54-729
|
5c29cd384a
|
调整num_consumed_batches在driver save中的逻辑
|
2022-04-14 16:04:12 +00:00 |
|
yh_cc
|
a4b2e0fac5
|
修复若干bug
|
2022-04-15 00:01:29 +08:00 |
|
YWMditto
|
2924e2117f
|
删除了 driver 中的 **_step,使用 model_call 和 get_model_call_fn 来代替;删除了 driver 中的所有 dataloaders
|
2022-04-14 23:34:35 +08:00 |
|
MorningForest
|
4cad7f548d
|
Merge remote-tracking branch 'refs/remotes/origin/dev0.8.0' into dev0.8.0
|
2022-04-14 19:37:02 +08:00 |
|
MorningForest
|
4be26c5620
|
修改torch fdl
|
2022-04-14 19:36:50 +08:00 |
|
yh_cc
|
8a0ffd6278
|
Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0
|
2022-04-14 16:02:55 +08:00 |
|
yh_cc
|
16a467393c
|
1.montior允许传入callable的对象进行选择; 2.解决Sampler中存在的循环引用问题
|
2022-04-14 16:02:41 +08:00 |
|
x54-729
|
17abdb12f6
|
Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0
|
2022-04-14 08:01:41 +00:00 |
|
x54-729
|
7c6e8b20a8
|
small
|
2022-04-14 08:01:14 +00:00 |
|
x54-729
|
64fa182aeb
|
修改断点重训部分逻辑
|
2022-04-14 08:01:04 +00:00 |
|
YWMditto
|
1452aa8f6c
|
修复了 dist 为 None 时的 set_dist_repro_dataloader 的逻辑
|
2022-04-14 13:50:53 +08:00 |
|
x54-729
|
ca93fccf62
|
Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0
|
2022-04-13 16:57:20 +00:00 |
|
YWMditto
|
2f23d80ccc
|
修改了 trainer 中的 validate 的调用的逻辑
|
2022-04-14 00:45:17 +08:00 |
|
YWMditto
|
8c0b0b8cd0
|
Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0
|
2022-04-13 19:09:34 +08:00 |
|
YWMditto
|
b9b0b53430
|
将 Events 修改为小写
|
2022-04-13 19:09:27 +08:00 |
|
x54-729
|
2b763138d0
|
Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0
|
2022-04-13 09:06:13 +00:00 |
|
x54-729
|
d2439fe443
|
修复_MetricsWrapper update传参的bug
|
2022-04-13 09:05:21 +00:00 |
|
yh_cc
|
31c6d5d02a
|
merge
|
2022-04-13 17:05:16 +08:00 |
|
yh_cc
|
9d71170bef
|
解决Trainer在断点重训的时候无法实现准确load和保存的问题
|
2022-04-13 17:04:33 +08:00 |
|
x54-729
|
f87723e2eb
|
small
|
2022-04-13 08:40:26 +00:00 |
|
YWMditto
|
1ac9e75c50
|
Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0
|
2022-04-13 15:37:15 +08:00 |
|
YWMditto
|
3ee6fc66f5
|
添加了 on_after_optimizers_step 和 on_after_zero_grad 的callback接口
|
2022-04-13 15:37:08 +08:00 |
|
x54-729
|
c2575ab357
|
Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0
|
2022-04-13 07:33:40 +00:00 |
|
x54-729
|
3ab93b2fae
|
paddle driver单卡和utils的pytest测试,添加了断点重训的测试
|
2022-04-13 07:33:27 +00:00 |
|
YWMditto
|
eb8d761a05
|
Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0
|
2022-04-13 14:27:08 +08:00 |
|
YWMditto
|
76a1e69022
|
little change
|
2022-04-13 14:27:01 +08:00 |
|
x54-729
|
5acaeabae4
|
Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0
|
2022-04-13 05:27:09 +00:00 |
|
yh_cc
|
e8d11cd5a9
|
1. 修复torch 分布式在不同版本中group参数default值不一样的问题; 2. torch修复多卡时只有batchsampler evaluate会遇到bug的问题; 3。logger增加warning_once接口;4.增加callback相关文档
|
2022-04-13 12:55:28 +08:00 |
|
x54-729
|
c61f28ce8e
|
Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0
|
2022-04-12 20:48:47 +00:00 |
|
YWMditto
|
8c22d0b1f6
|
修改了 Trainer.on 的错误提示
|
2022-04-12 22:47:39 +08:00 |
|