Comments (13)
Hi, @lmohit95, this has been solved in #52 and sorry for the delay.
from hetu.
Thank you. load_data.py works perfectly now. The dataset is downloaded and processed.
But while running bash tests/local_dcn_criteo.sh
command, I get the following error. I have tried changing batchsize
at line 62 in run_hetu.py and tried running the command, but I still get the same error.
I am able to run other DLRM like facebook open source DLRM using GPU, so I believe CUDA setup is correct.
from hetu.
@lmohit95, Hetu main brench has been updated which enables dynamic memory, please pull the new code and try again.
from hetu.
@lmohit95, Hetu main brench has been updated, please try to pull the new code and try again.
Thank you. It works now. Sorry for asking lot of questions. I am facing this issue now while training criteo dataset.
from hetu.
@lmohit95, Hetu main brench has been updated, please try to pull the new code and try again.
Thank you. It works now. Sorry for asking lot of questions. I am facing this issue now while training criteo dataset.
I mean this problem has been solved by #47 which was merged not long before, and will it still happen when you pull these changes?
from hetu.
I get the following error when I run bash tests/local_dcn_criteo.sh
. I created hetu_config.yaml
file in tmp folder and copied contents provided in README.MD. I am trying to run HET on a single GPU.
To avoid this error, I deliberately made file = None
in __init__
function of distribute.py. While doing that, I am facing the outofmemory error.
from hetu.
@lmohit95, You can also update the line 120 by the following code to avoid the first error:
Lines 120 to 126 in 1684091
For the OOM error, #47 implements dynamic memory allocation, and the gpu memory peak will be halved when you run bash tests/local_dcn_criteo.sh
.
Maybe you haven't pull the latest code yet?
from hetu.
Thanks a lot for everything. The tests are working perfectly now. I was accessing the forked repo mentioned in the HET paper.
While running python run_hetu.py --model dcn_criteo --all --val
command, I am facing the following error:
I pulled latest code and downloaded criteo dataset by running load_data.py file.
from hetu.
@lmohit95, thanks for you feedback and sorry for my mistake.
It is true that there are still some errors in dataset processing, and I have fixed it in #54.
Please pull my code and run python load_data.py
again before running run_hetu.py --model dcn_criteo --all --val
.
from hetu.
Thank you. Everything works now. I just wanted to clarify something regarding run_hetu.py --model dcn_criteo --all --val
. This command trains and tests HET on criteo dataset right?
The paper mentions that the training process can take hours (Fig 6),
but in my case the training runs for a total of 10 epochs with far less overall runtime.
from hetu.
It seems like you are running in a local execution mode, rather than the distributed training. That's why it's much faster.
Besides, note that the test_auc is reported every 1/10 epoch as described in
Line 190 in acae42a
from hetu.
Got it. Thanks for all the help!!!
from hetu.
Hello,
Thanks for all the help until now.
I am running HET on criteo dataset on a single GPU node by setting HETU_VERSION = 'gpu' in HYBRID mode. I ran `bash examples/ctr/tests/hybrid_wdl_criteo.sh, but I am getting the following error:
This is my configuration file
shared :
DMLC_PS_ROOT_URI : 127.0.0.1
DMLC_PS_ROOT_PORT : 13100
DMLC_NUM_WORKER : 2
DMLC_NUM_SERVER : 1
DMLC_PS_VAN_TYPE : p3
launch :
worker : 2
server : 1
scheduler : true
nodes:
- host: lmohit95
servers: 1
workers: 2
chief: true
from hetu.
Related Issues (19)
- Duplicate ps when building Hetu HOT 8
- How to run ctr examples with GPU HOT 1
- Get stuck when launching the server in hybrid mode HOT 1
- run PS的demo时触发OSError HOT 1
- ModuleNotFoundError: No module named 'hetu_cache' HOT 1
- 是否考虑支持容器部署和编译 HOT 1
- PS Worker Sparse Push Block
- fatal error: cub/cub.cuh: No such file or directory HOT 2
- The Question about reverse_layout_transform_kernel HOT 1
- AssertionError: assert params_size == len(grads) HOT 1
- Invalid device ordinal error HOT 5
- Backward AllReduce in HET
- Error when running RDMA HOT 4
- I can't find source code about Automatic Distributed Training, i.e. Cost estimation HOT 3
- 我想问下Hetu能部署在国产显卡上吗,比如DCU和Ascend HOT 1
- Questions about some specific numbers in Galvatron. HOT 4
- Questions about moe examples.
- Question about using galvatron AssertionError: 50257 is not divisible by 4 HOT 1
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from hetu.