profile
viewpoint

frogcjn/AdaptivePopover_iOS8_Swift 26

Adaptive Popover using UIPopoverPresentationController, which is new feature in iOS8

frogcjn/UICollectionView-Move-Drag-Item-iOS9 23

This sample code shows how to use official API to move/drag item in collection, which is new feature in iOS9.

frogcjn/OffsetIndexableCollection-String-Int-Indexable- 5

OffsetIndexableCollection (String Int Indexable)

frogcjn/iPhoneX-Lock-Home-Indicator 4

This is an example for demoing how to use official API of edge protection to lock the home indicator for games and special apps.

frogcjn/vscode-ReactNative-HelloWorld 4

This is the hello world project for using React Native in Visual Studio Code

frogcjn/ImageIOPlus 2

ImageIOPlus Swift Package

frogcjn/ArrayProtocol-ArrayWrapper 1

This repo contains the ArrayLikeProtocol and ArrayProtocol, and an ArrayWrapper to emulate the Array in Swift. Thus you can subclass an "Array" easily.

frogcjn/vscode-react-native 1

vscode extension for react native

frogcjn/DefinitelyTyped 0

The repository for high quality TypeScript type definitions.

push eventTsinghuaAI/CPM-Generate

Zhengyan Zhang

commit sha acc91e1ad8813e8f8fd813a3511e77593eb362b4

Update README.md

view details

push time in 2 hours

push eventTsinghuaAI/CPM-Generate

Zhengyan Zhang

commit sha 35d8a14adc06a79fa23dac5dd12489c4e2fb23c2

Update README.md

view details

push time in 2 hours

issue commentTsinghuaAI/CPM-Generate

结果无法复现

@leecuijiao 会一定程度影响结果,拼接方式就是推送里写的那样,每个输入一行。另外,模型的生成结果是随机采样的,所以也没法保证每次都生成一样的。

leecuijiao

comment created time in 3 hours

issue closedTsinghuaAI/CPM-Generate

2路模型并行的参数,转换到4路模型?

你好,我想基于这模型在自己的数据集做finetuing,但是在3090上只能将模型分在4个gpu训练,无法像在预训练那样将模型分到2个gpu,所以需要将参数合并,并在某些层均分层4份,并保存,这种做法可以吗?

closed time in 5 hours

bigprince97

issue commentTsinghuaAI/CPM-Generate

结果无法复现

@leecuijiao 在运行generate_text.sh时我遇到这个错误?请问你有遇到过吗?怎么解决的?谢谢! Generate Samples Generate Samples WARNING: No training data specified WARNING: No training data specified using world size: 2 and model-parallel size: 2

using dynamic loss scaling Traceback (most recent call last): File "/opt/conda/lib/python3.6/runpy.py", line 193, in _run_module_as_main "main", mod_spec) File "/opt/conda/lib/python3.6/runpy.py", line 85, in _run_code exec(code, run_globals) File "/opt/conda/lib/python3.6/site-packages/torch/distributed/launch.py", line 253, in main() File "/opt/conda/lib/python3.6/site-packages/torch/distributed/launch.py", line 249, in main cmd=cmd) subprocess.CalledProcessError: Command '['/opt/conda/bin/python', '-u', 'generate_samples.py', '--local_rank=1', '--model-parallel-size', '2', '--num-layers', '32', '--hidden-size', '2560', '--load', '/gdata/ChineseCorpus/CPM-model-v1', '--num-attention-heads', '32', '--seq-length', '1024', '--max-position-embeddings', '1024', '--fp16', '--cache-dir', 'cache', '--out-seq-length', '512', '--temperature', '0.9', '--top_k', '0', '--top_p', '0', '--tokenizer-path', 'bpe_3w_new/', '--vocab-size', '30000']' died with <Signals.SIGSEGV: 11>.

@zzy14

请问你解决了吗,我也碰到了你这个问题

leecuijiao

comment created time in a day

issue openedTsinghuaAI/CPM-Generate

请问怎么转换成在一个GPU上运行的模型呢?

需要修改哪些代码及如何转换模型呢?

created time in a day

issue openedTsinghuaAI/CPM-Generate

请问可以在windows上运行吗

请问可以在windows上运行吗

created time in a day

issue commentTsinghuaAI/CPM-Generate

2路模型并行的参数,转换到4路模型?

转化成功了,谢谢

bigprince97

comment created time in 2 days

issue commentTsinghuaAI/CPM-Generate

结果无法复现

@zzy14 拼接方式会影响结果吗,汇报中的demo是使用的什么拼接方式呢

leecuijiao

comment created time in 2 days

issue closedTsinghuaAI/CPM-Generate

NCCL Runtime Error

你好,我在运行您代码的时候出现了NCCL错误的问题。我的环境是 pytorch/pytorch:1.7.0-cuda11.0-cudnn8-devel的镜像创建的容器。网上查了一些解决方案并不得行,请问您能否发布一个能顺利运行的docker镜像 【错误提示】 RuntimeError: NCCL error in: /opt/conda/conda-bld/pytorch_1603729096996/work/torch/lib/c10d/ProcessGroupNCCL.cpp:784, unhandled system error, NCCL version 2.7.8

closed time in 2 days

JJack0812

issue commentTsinghuaAI/CPM-Generate

NCCL Runtime Error

啊我太蠢了,这是torch版本太高引发的错误。的确要用1.6.0的

JJack0812

comment created time in 2 days

push eventTsinghuaAI/CPM-Generate

Zhengyan Zhang

commit sha 95c9761edd224bc210ee63f060e90f106e3cc645

Update README.md

view details

push time in 3 days

push eventTsinghuaAI/CPM-Generate

zzy

commit sha e3afd700d3dfbd1fc6155bc5eb0fd97d0f3e6842

Simplify tokenization.

view details

push time in 3 days

issue commentTsinghuaAI/CPM-Generate

结果无法复现

@FannierPeng 我也没遇到过你这个错,我们会在近期准备一个docker。

leecuijiao

comment created time in 3 days

issue commentTsinghuaAI/CPM-Generate

结果无法复现

@leecuijiao 不需要微调,我们汇报的例子是来源于实现的一个demo,demo的是现实随机生成多个预测结果,然后每个预测结果在换行符、句号、逗号进行阶段。

leecuijiao

comment created time in 3 days

issue commentTsinghuaAI/CPM-Generate

2路模型并行的参数,转换到4路模型?

理论上是可以的,按照Megatron的并行思路,每一层都要均分成4份,这一点你可能需要注意一下。

bigprince97

comment created time in 3 days

issue commentTsinghuaAI/CPM-Generate

结果无法复现

@leecuijiao 在运行generate_text.sh时我遇到这个错误?请问你有遇到过吗?怎么解决的?谢谢!

Generate Samples Generate Samples WARNING: No training data specified WARNING: No training data specified using world size: 2 and model-parallel size: 2

using dynamic loss scaling Traceback (most recent call last): File "/opt/conda/lib/python3.6/runpy.py", line 193, in _run_module_as_main "main", mod_spec) File "/opt/conda/lib/python3.6/runpy.py", line 85, in _run_code exec(code, run_globals) File "/opt/conda/lib/python3.6/site-packages/torch/distributed/launch.py", line 253, in main() File "/opt/conda/lib/python3.6/site-packages/torch/distributed/launch.py", line 249, in main cmd=cmd) subprocess.CalledProcessError: Command '['/opt/conda/bin/python', '-u', 'generate_samples.py', '--local_rank=1', '--model-parallel-size', '2', '--num-layers', '32', '--hidden-size', '2560', '--load', '/gdata/ChineseCorpus/CPM-model-v1', '--num-attention-heads', '32', '--seq-length', '1024', '--max-position-embeddings', '1024', '--fp16', '--cache-dir', 'cache', '--out-seq-length', '512', '--temperature', '0.9', '--top_k', '0', '--top_p', '0', '--tokenizer-path', 'bpe_3w_new/', '--vocab-size', '30000']' died with <Signals.SIGSEGV: 11>.

@zzy14

leecuijiao

comment created time in 4 days

issue commentTsinghuaAI/CPM-Generate

结果无法复现

@leecuijiao 在运行generate_text.sh时我遇到这个错误?请问你有遇到过吗?怎么解决的?谢谢!


Generate Samples Generate Samples WARNING: No training data specified WARNING: No training data specified using world size: 2 and model-parallel size: 2

using dynamic loss scaling Traceback (most recent call last): File "/opt/conda/lib/python3.6/runpy.py", line 193, in _run_module_as_main "main", mod_spec) File "/opt/conda/lib/python3.6/runpy.py", line 85, in _run_code exec(code, run_globals) File "/opt/conda/lib/python3.6/site-packages/torch/distributed/launch.py", line 253, in <module> main() File "/opt/conda/lib/python3.6/site-packages/torch/distributed/launch.py", line 249, in main cmd=cmd) subprocess.CalledProcessError: Command '['/opt/conda/bin/python', '-u', 'generate_samples.py', '--local_rank=1', '--model-parallel-size', '2', '--num-layers', '32', '--hidden-size', '2560', '--load', '/gdata/ChineseCorpus/CPM-model-v1', '--num-attention-heads', '32', '--seq-length', '1024', '--max-position-embeddings', '1024', '--fp16', '--cache-dir', 'cache', '--out-seq-length', '512', '--temperature', '0.9', '--top_k', '0', '--top_p', '0', '--tokenizer-path', 'bpe_3w_new/', '--vocab-size', '30000']' died with <Signals.SIGSEGV: 11>.


leecuijiao

comment created time in 4 days

issue openedTsinghuaAI/CPM-Generate

2路模型并行的参数,转换到4路模型?

你好,我想基于这模型在自己的数据集做finetuing,但是在3090上只能将模型分在4个gpu训练,无法像在预训练那样将模型分到2个gpu,所以需要将参数合并,并在某些层均分合并,并保存,这种做法可以吗?

created time in 4 days

issue openedTsinghuaAI/CPM-Generate

NCCL Runtime Error

你好,我在运行您代码的时候出现了NCCL错误的问题。我的环境是 pytorch/pytorch:1.7.0-cuda11.0-cudnn8-devel的镜像创建的容器。网上查了一些解决方案并不得行,请问您能否发布一个能顺利运行的docker镜像 【错误提示】 RuntimeError: NCCL error in: /opt/conda/conda-bld/pytorch_1603729096996/work/torch/lib/c10d/ProcessGroupNCCL.cpp:784, unhandled system error, NCCL version 2.7.8

created time in 4 days

issue commentTsinghuaAI/CPM-Generate

TypeError: Class advice impossible in Python3. Use the @implementer class decorator instead.

linux python3.6也出现了这个问题。

zhaogangthu

comment created time in 4 days

issue openedTsinghuaAI/CPM-Generate

结果无法复现

请问这篇智源研究院发布的文章 https://mp.weixin.qq.com/s/QinmwLIRfYL9N2LfoFoEBQ 中提到的样例展示的第一个,用模型预测的结果不对,请问需要怎么拼接或者需要微调吗 image

created time in 5 days

issue openedTsinghuaAI/CPM-Generate

TypeError: Class advice impossible in Python3. Use the @implementer class decorator instead.

from apex.normalization.fused_layer_norm import FusedLayerNorm as LayerNorm

请问是python版本的问题吗?

created time in 5 days

fork a710128/concourse-rsync-resource

concourse.ci resource for persisting build artifacts on a shared storage location with rsync and ssh.

fork in 6 days

issue closedTsinghuaAI/CPM-Generate

有开源训练代码的打算吗

想做一些下游任务的微调,如有训练代码就好啦

closed time in 6 days

Line290

issue commentTsinghuaAI/CPM-Generate

有开源训练代码的打算吗

好的,昨天根据deepspeed的例子改了一下训练代码,应该可以finetune了

Line290

comment created time in 6 days

issue commentTsinghuaAI/CPM-Generate

有开源训练代码的打算吗

请参考 #8.

Line290

comment created time in 6 days

issue commentTsinghuaAI/CPM-Generate

如何进行微调?没有详细文档,也没有更详细的使用例子?

我们在实验里主要验证了少次学习的能力(in-context learning),finetune的代码我们会后续整理,这个repo主要用于文本生成。

xiaomail0991

comment created time in 6 days

more