profile
viewpoint
If you are wondering where the data of this site comes from, please visit https://api.github.com/users/Sleepychord/events. GitMemory does not store any data, but only uses NGINX to cache data for a period of time. The idea behind GitMemory is simply to give users a better reading experience.

THUDM/cogdl 808

CogDL: An Extensive Toolkit for Deep Learning on Graphs

Sleepychord/ImprovedGAN-pytorch 254

Semi-supervised GAN in "Improved Techniques for Training GANs"

Sleepychord/CogLTX 131

The source code of NeurIPS 2020 paper "CogLTX: Applying BERT to Long Texts"

THUDM/GraphSGAN 67

Implementation of "GraphSGAN", a GAN-based semi-supervised learning algorithm for graph data.

THUDM/HGB 40

Revisiting, benchmarking, and refining Heterogeneous Graph Neural Networks.

Sleepychord/ChinesePoemGeneration 35

基于深度学习的古诗生成(deep learning based Chinese poem generation)

Sleepychord/pboc-arduino 8

Arduino PBOC Library using MFRC522

luogan1234/-The_King_of_Fighters 4

digital logic design course's big project

Sleepychord/GAN-paper-reading 4

GAN相关论文的中文笔记

Sleepychord/Poisson-Image-Editing 3

implementation of classic paper "Poisson Image Editing"

push eventSleepychord/cogdata

root

commit sha 33c7868ff3ead0fe4bf192be74fa5084b6a80fb6

reverse save order

view details

push time in 11 days

issue commentTHUDM/CogView

docker pull error, "You have reached your pull rate limit"

Hi, the dockerfile is already under env/, (although I am not 100% sure if it is the correct version).

XuanxuanGao

comment created time in 18 days

push eventSleepychord/Image-Local-Attention

Ming Ding

commit sha 88eb3ea4f12c9052867037d0c2ffb96ea0364033

test before cast

view details

Ming Ding

commit sha 2ef4334584fccc3c0cf999509de8593b78e26088

Merge branch 'master' of github.com:Sleepychord/Image-Local-Attention

view details

Ming Ding

commit sha 9948fdaa41c8dcdc17662534cbec01eb9ab9b64d

grad contiguous

view details

push time in a month

issue commentTHUDM/CogView

The math in this paper

“因为我们在假设的生成过程的step(3)中,认为z已经完全拥有了文本的信息”--這裡我說的是隨機變量圖模型的“生成過程”,這裡是follow了VAE原文的說法,並不是指訓練過程。訓練過程第一階段沒有圖像參與,你可以理解為圖像的一個壓縮的過程。

如果沒有見到大象,顯然是不可能生成出大象了……但是可能學會組合,比如紅色+老虎之類的。

brianw0924

comment created time in a month

issue closedTHUDM/CogView

Layernorm form in paper

The formulation of layernorm in paper multiply a root_square(d) compare to layernorm introduced in pytorch document. Why add this multiplication? thank you image

closed time in a month

mzy97

issue commentTHUDM/CogView

Layernorm form in paper

No, it is the original LayerNorm. Note d is the length of the vector x.

mzy97

comment created time in a month

issue commentTHUDM/CogView

The demo website is broken?

改成简体中文似乎是可以的。

brianw0924

comment created time in a month

issue commentTHUDM/CogView

The math in this paper

1.加入文字之後的ELBO推導與之前無異,但是注意其他部分增加一個given t_i,并预先分离出文字部分的NLL,便可得到公式(2)。

brianw0924

comment created time in a month

issue closedTHUDM/CogView

The demo website is broken?

請問為什麼那個 demo 網站輸入任何文字都說有非法內容

closed time in a month

brianw0924

issue commentTHUDM/CogView

The demo website is broken?

非法内容指“色情、暴力、歧视、辱骂和政治敏感内容”,由于有人报告demo被用于不良用途,我们加大了关键字的判别力度,因此可能误伤正常的查询,非常抱歉。https://github.com/THUDM/CogView/issues/13

brianw0924

comment created time in a month

issue closedTHUDM/CogView

script to finetune Cogview-base

Hi, I'm trying to finetune Cogview pretrained model. However, when I try to load model weights, I get following error: RuntimeError: Error(s) in loading state_dict for GPT2Model:
size mismatch for word_embeddings.weight: copying a param with shape torch.Size([14560, 2560]) from checkpoint, the shape in current model is torch .Size([14592, 2560]).

Here is my script:

`NUM_WORKERS=1 NUM_GPUS_PER_WORKER=4 MP_SIZE=1

script_path=$(realpath $0) echo $script_path script_dir=$(dirname $script_path) main_dir=$(dirname $script_dir)

OPTIONS_NCCL="NCCL_DEBUG=info" HOST_FILE_PATH="hostfile_single"

config_json="$script_dir/ds_config_zero.json" gpt_options="
--experiment-name cogview-test_finetune
--img-tokenizer-num-tokens 8192
--dataset-type TokenizedDataset
--model-parallel-size ${MP_SIZE}
--batch-size 4
--num-layers 48
--hidden-size 2560
--num-attention-heads 40
--save ./
--train-iters 2000
--save-interval 800
--resume-dataloader
--train-data /path/to/my/data
--split 90,5,5
--distributed-backend nccl
--lr-decay-style cosine
--warmup .1
--checkpoint-activations
--deepspeed-activation-checkpointing
--max-position-embeddings 1089
--max-memory-length 0
--fp16
--txt-loss-scale 5
--load /path/to/cogview
--no-load-rng
--model-parallel-size 2
--num-workers 16
--is-sparse 0
--finetune
--shuffle "

gpt_options="${gpt_options} --deepspeed
--deepspeed_config ${config_json}
"

run_cmd="${OPTIONS_NCCL} deepspeed --num_nodes ${NUM_WORKERS} --num_gpus ${NUM_GPUS_PER_WORKER} --hostfile ${HOST_FILE_PATH} pretrain_gpt2.py $@ ${gpt_options}" `

It will be great if you can provide some details for finetuning. Thanks!

closed time in a month

luyang-huang96

issue commentTHUDM/CogView

script to finetune Cogview-base

Hi, It depends on your memory and total number of machines (if you are using ZeRO), so the best way may be to try by yourself.

luyang-huang96

comment created time in a month

push eventSleepychord/Image-Local-Attention

Ming Ding

commit sha ce164bd6554bc2b5b99da260f19ca2c5e947b348

neg

view details

push time in a month

issue commentTHUDM/HGB

在线提交预测结果出错

提交文件是预测结果文件的压缩包,提交格式可以参考:https://www.biendata.xyz/competition/hgb-1/evaluation/

每次提交四个数据集,一共二十个文件的话,这二十个文件的score都会有吗,还是一个数据集一个score,或者这四个数据集只有一个score

提交之后在选手的提交记录中可以知道分别的score,但是biendata比赛页面只能显示总的,你提交一下试试就知道了。

wjcPure

comment created time in a month

push eventSleepychord/Image-Local-Attention

Ming Ding

commit sha 868c97977b68b26c06d64279e9127a577b799034

large to small resolution

view details

push time in a month

issue closedTHUDM/CogQA

Where can i get `hotpot_train_v1.1.json`?

When i run !python /content/CogQA/process_train.py the following error occures:

Traceback (most recent call last): File "/content/CogQA/process_train.py", line 18, in <module>

with open('./hotpot_train_v1.1.json', 'r') as fin:

FileNotFoundError: [Errno 2] No such file or directory: './hotpot_train_v1.1.json'

I wonder where can i get this file, please help

closed time in a month

sexozavr

issue commentTHUDM/CogQA

Where can i get `hotpot_train_v1.1.json`?

https://hotpotqa.github.io/

sexozavr

comment created time in a month

push eventSleepychord/Image-Local-Attention

Ming Ding

commit sha b39ebbcfee1e1b1af15f83bb4fdeeec7c18442ab

add auto expand for different batch

view details

push time in a month

push eventSleepychord/Image-Local-Attention

Ming Ding

commit sha e9d7861d167acf7b8cad6019d486dc44e5a49837

speed up, fp16 and casual_mask

view details

push time in a month

issue commentTHUDM/CogView

script to finetune Cogview-base

Hi, could you change the model_parallel_size back to 1 and try again? our released model use model_parallel_size=1

luyang-huang96

comment created time in a month

push eventTHUDM/CogView

Sleepy_chord

commit sha 8abd84d6fa91bd2cf03d427eb3cd8a1487213147

add pointer to cogdata

view details

push time in a month

issue closedTHUDM/CogView

vqvae pretrained model

Hi, thank you for your excellent work? How can I train my own VQVAE model? or VQGAN model?

closed time in a month

ToBeNormal

issue commentTHUDM/CogView

vqvae pretrained model

Hi, we have already update another the example dataset, which is only about 80,000 images. The training performance should not be as good as the released CogView. It is only an example dataset to try.

ToBeNormal

comment created time in a month

issue closedTHUDM/CogView

How long will it take for a pre-training model application to be approved?

@neozhangthe1 @Sleepychord @lykeven @cenyk1230

closed time in a month

starmemda

issue closedTHUDM/CogView

Hi! How should I prepare the dataset_type of TextCodedataset in your code?

Hi! How should I prepare the dataset_type of TextCodedataset in your code if I want to finetune the network in our data. I notice that TextCodedataset consists of text and code. What does the code mean? And how should I get it? QQ20210726-100221 @neozhangthe1 @Sleepychord @lykeven @cenyk1230 @Somefive

closed time in a month

starmemda

issue commentTHUDM/CogView

Hi! How should I prepare the dataset_type of TextCodedataset in your code?

Hi, we have also add another data format, BinaryDataset. It is only a series of 64 text tokens and 1024 image tokens. You can easily generate such file yourself or by our cogdata toolkit.

starmemda

comment created time in a month

issue closedTHUDM/CogView

the Alibaba item-title image tokens dataset from our link at Tianchi(TODO)

the link to the Alibaba item-title image tokens dataset at Tianchi(TODO) seems to be unavailable

closed time in a month

WXLL579

issue commentTHUDM/CogView

the Alibaba item-title image tokens dataset from our link at Tianchi(TODO)

Hi, we have already update another the example dataset, which is only about 80,000 images. The training performance should not be as good as the released CogView. It is only an example dataset to try.

WXLL579

comment created time in a month

push eventTHUDM/CogView

Ming Ding

commit sha 7b90b24c2abafcf3b4195b6d3164be2b21d64fc4

add compact bin and bird animal example dataset

view details

push time in a month

startedbluenote10/yachalk

started time in a month