x54-729
|
9b13fd313a
|
small
|
2022-04-16 08:39:38 +00:00 |
|
x54-729
|
3dbb3677f0
|
微调 reproducible sampler 的初始化
|
2022-04-16 08:39:07 +00:00 |
|
x54-729
|
aa2f678507
|
Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0
|
2022-04-16 07:23:18 +00:00 |
|
yh_cc
|
d813f31f9a
|
1.删除sampler中num_consumed_samples_array,转为通过batch_size等进行换算。2.在Metric中新增all_gather_object接口,方便评测
|
2022-04-16 15:22:44 +08:00 |
|
x54-729
|
6bfdb39c2f
|
完善paddle fleet set_dist_repro_dataloader的测试例
|
2022-04-16 06:42:29 +00:00 |
|
x54-729
|
fcd27cfc3f
|
添加FASTNLP_NO_SYNC相关的设置
|
2022-04-16 05:50:53 +00:00 |
|
x54-729
|
a25a73394b
|
Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0
|
2022-04-16 05:40:35 +00:00 |
|
x54-729
|
de544707d9
|
paddle fleet set_dist_repro_dataloader的测试例
|
2022-04-16 05:40:23 +00:00 |
|
yh_cc
|
d758a36d4a
|
update metric的实现
|
2022-04-16 12:40:29 +08:00 |
|
yh_cc
|
048e409233
|
Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0
|
2022-04-16 00:33:32 +08:00 |
|
yh_cc
|
aaa4ccc3e0
|
1.Metric的aggregate_when_get_metric默认值修改为None;将根据Evaluator中Sampelr的状态最终决定;2.增加fastnlp_bo_syn_context关闭fastnlp中的多卡同步操作
|
2022-04-16 00:32:52 +08:00 |
|
x54-729
|
32d8e27472
|
small
|
2022-04-15 16:10:10 +00:00 |
|
x54-729
|
26c80d620c
|
paddle单卡加载fp16的测试
|
2022-04-15 16:01:21 +00:00 |
|
x54-729
|
8cda30c426
|
small
|
2022-04-15 15:10:35 +00:00 |
|
x54-729
|
5d1ac72ec9
|
加载fp16时同时设置auto_cast和fp16属性
|
2022-04-15 14:38:29 +00:00 |
|
x54-729
|
262bc1a82e
|
small
|
2022-04-15 14:03:08 +00:00 |
|
x54-729
|
8e9a47cf00
|
Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0
|
2022-04-15 13:48:23 +00:00 |
|
MorningForest
|
a2956b697e
|
Merge branch 'dev0.8.0' of github.com:fastnlp/fastNLP into dev0.8.0
|
2022-04-15 20:04:53 +08:00 |
|
MorningForest
|
665d79a3ed
|
增加paddle单卡的accuracy测试用例
|
2022-04-15 20:03:44 +08:00 |
|
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 |
|