Commit Graph

2472 Commits

Author SHA1 Message Date
x54-729
8e0d03a3d1 修改 Trainer torch_kwargs paddle_kwargs fairscale_kwargs 的描述 2022-06-28 23:24:16 +08:00
x54-729
142fd47601 修改 TorchDataLoader PaddleDataLoader set_pad 的描述 2022-06-28 01:06:05 +08:00
x54-729
513a5f875c update paddle tutorial_e1 2022-06-27 19:08:48 +08:00
x54-729
dabb8b9785 paddle tutorial 情感分析改名 2022-06-27 19:07:56 +08:00
x54-729
8b1ed86033 fix paddle dataset __getattr__ 2022-06-25 21:25:22 +08:00
x54-729
6e89cb77c3 Evaluator在结果为空时不会进行输出 2022-06-22 23:39:53 +08:00
x54-729
9e3043251a fastnlp paddle ernie 的 tutorial 2022-06-22 23:39:33 +08:00
x54-729
78596ea11c 为 Trainer 的driver 参数增加 'auto' 选项 2022-06-22 16:49:03 +08:00
yhcc
eb0e563fec 修复OverfitBatches被替换的问题 2022-06-22 16:35:32 +08:00
x54-729
0e865d292d Merge remote-tracking branch 'origin/dev0.8.0' into deepspeed 2022-06-20 20:53:34 +08:00
x54-729
44d2a574ae logger.warn->logger.warning 2022-06-20 20:52:53 +08:00
x54-729
6f9d703f13 logger.warn->logger.warning 2022-06-20 20:52:04 +08:00
x54-729
a495bb938a 1.修复模型会被移动到rank对应设备的问题 2.更改 deepspeed driver æ˜命名 3.为 deepspeed 添加 logging_level 2022-06-20 20:49:17 +08:00
yhcc
de5d5597e7 修复fitlogcallback增加launch_time的bug 2022-06-20 12:58:42 +08:00
x54-729
2735d2d10c DeepSpeedDriver现在可以通过 deepspeed 命令拉起;添加了相关 trainer 的简单测试 2022-06-20 06:34:25 +08:00
x54-729
8d23253318 import DeepSpeedDriver 2022-06-20 06:33:14 +08:00
x54-729
7023ea550c deepspeed的save load功能 2022-06-20 02:26:34 +08:00
yhcc
8467cb6e41 fix bug and make shuffle automatic 2022-06-19 17:16:39 +08:00
x54-729
687651e35f Merge remote-tracking branch 'origin/dev0.8.0' into deepspeed 2022-06-18 23:05:32 +08:00
x54-729
9903d2eec1 TorchDriver的sampler加载和保存拆分为单独的函数 2022-06-18 23:04:50 +08:00
x54-729
2d2bf421fd deepspeed checkpoint相关函数(ï微未测试) 2022-06-18 22:28:57 +08:00
x54-729
22d95be007 Merge branch 'dev0.8.0' into deepspeed 2022-06-17 23:31:55 +08:00
x54-729
b60621f3d1 small 2022-06-17 23:23:43 +08:00
x54-729
64da46b613 paddle replace_batch_sampler和check_dataloader 跟进 2022-06-17 23:23:33 +08:00
x54-729
d26d0ad17f 添加选择deepspeed driver的逻辑 2022-06-17 22:16:55 +08:00
x54-729
cbee1c6cbc deepspeed test init 2022-06-17 22:12:38 +08:00
x54-729
a39d3011e8 Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0 2022-06-17 21:59:04 +08:00
x54-729
dca3377129 ddp添加环境变量RANK的设置 2022-06-17 21:58:50 +08:00
x54-729
1a2eb93ab4 deepspeed基本功能 2022-06-17 21:55:59 +08:00
YWMditto
bb68856f85 添加对 overfit 多卡的测试 2022-06-17 00:53:13 +08:00
yhcc
399065ae04 update overfit_batches 2022-06-17 00:12:53 +08:00
YWMditto
a6fc5225cd 添加了 overfit_batches 的注释 2022-06-16 22:37:43 +08:00
YWMditto
024fecfbf3 添加了 overfit 的功能 2022-06-16 22:30:02 +08:00
x54-729
55d8738def deepspeed driver init 2022-06-14 15:03:41 +00:00
x54-729
cd0957fb5b 跟进paddle jittor 关于 set_dist_repro_dataloader函数中的修改 2022-06-14 13:19:56 +00:00
yhcc
dd1c5ca035 修复设置了global seed的bug 2022-06-13 23:15:06 +08:00
yhcc
7283bf27b2 在私有定制Sampler的情况下,多卡不替换 2022-06-13 20:57:25 +08:00
YWMditto
70dea71cdb 为 ddp 在用户使用自己的sampler 和 batch sampler 是添加禁止 2022-06-13 17:53:23 +08:00
yhcc
e4a7e64600 progress打印增加一种特殊yueding 2022-06-11 22:30:20 +08:00
yhcc
0bf0dab347 增加一个TimerCallback用于计时 2022-06-07 14:52:25 +08:00
x54-729
c99315f79e 修复新增的set_dist_repro_dataloader函数测试例在paddle情况下的问题 2022-06-05 23:55:38 +00:00
yhcc
47b7c4e832 增加set_dist_repro_dataloader测试 2022-06-05 23:15:51 +08:00
YWMditto
8d0602f5de 修复 torch_driver/utils/replace_sampler 的bug 2022-06-05 20:07:36 +08:00
yhcc
08416a3a6c 1.修复lr_schedulder的调用时机问题;2.修复replace_sampler的初始化问题 2022-06-05 19:51:44 +08:00
MorningForest
bd7c6abd32 Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0 2022-06-05 15:52:18 +08:00
MorningForest
3e2960627a acc 2022-06-05 15:52:09 +08:00
lxr-tech
8319706f02 finish tutorial-3456 lxr 220604 2022-06-04 21:15:44 +08:00
x54-729
4cc8a29926 1.为element测试添加torch标签 2.解决torch版本导致的int张量无法对整数使用除法的问题 2022-06-04 08:27:42 +00:00
yhcc
24092e3114 增加pipe相关的测试 2022-06-04 15:22:22 +08:00
MorningForest
637919e45d Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0 2022-06-04 15:09:11 +08:00