Some issues regarding the execution of scripts #9
Comments
|
Did you follow steps mentionned in readme.md ? |
|
Hi, thank you for using our code ! We do not have enough information to help you with your problems. Please provide us the following:
Thanks ! |
|
Before starting anything, I have installed : Python 3.6.2, Java (openjdk 10), Maven, PyTorch (working with CUDA, checked), and the model packages (Allen Bert & Transformers). I've also downloaded the repository disambiguate, the repository UFSAC, and the link of the data (DATADIR). Then, I compiled using mvn in the UFSAC repo, then I launched compile.sh from disambiguate repo :
Then, I tried to run decode.sh this way : In the case of evaluate.sh, I ran the following command and I got this output (I have tried to change the corpus parameter but it didn't change anything) : ------ Evaluate the score of an ensemble of models Evaluate on corpus ../ufsac-public-2.1/semcor.xml ------ Evaluate the scores of individual models Evaluate on corpus ../ufsac-public-2.1/semcor.xml PS : Sorry for the late answer, and thank you very much for both replies! :) |
|
Hi, First, about your issue on the About your second error, this is really weird... It seems that the Python process launched by the Java code is killed during the evaluation/decode process.... Can you run the following command: |
|
Hello,
[jamilz@jamil-pc disambiguate-master]$ cat foo [jamilz@jamil-pc disambiguate-master]$ ./decode.sh --data_path ../data/ --weights ../data/model_weights_wsd0 < foo
[jamilz@jamil-pc python]$ ./launch.sh getalp.wsd.predict --data_path ../../data/ --weights ../../data/model_weights_wsd0 --disambiguate It seems like, as you said, the Python process keep getting killed in every case, I am currently trying to figure out why it does that. |
|
Indeed, this problem takes place, since I have an exactly same output as in previous comment when running decode.sh script. |
|
But when I run the script from python directory (as suggested above), I get the following output: (env) andry-bal@andrybal-VirtualBox:~/Documents/NLP/disambiguate/python$ ./launch.sh getalp.wsd.predict --data_path ../../data --weights ../../data/model_weights_wsd0 --disambiguate |
|
@Andry-Bal Thank you, it gives more informations for debugging :) So it seems that the "weights" file is somewhat corrupted or something like that. Is your file located at |
|
The file located in I re-downloaded it in order to check if it is because of file corruption. It didn't change anything. In my opinion, it might be because of my computer because, as the process stops working in the middle of execution, it can be a problem with memory. I don't see what kind of other problem would kill a process like that. |
|
Oh, yes you're right, it's probably a lack of memory, because the provided model is really large. My advice is that if you use your personal computer, you should train a new model by yourself with a smaller BERT model (like Albert or Distilbert). I'll keep the issue open to remind me to catch the error and try to display a more explicit message. |
|
@coylz I have doubts, that this is a memory issue. I monitored RAM and disk during script execution and did not see anything suspicious (however I might be mistaken). |
|
@Andry-Bal Thank you for comment, it will help both us and other users. |
|
Thank you very much @coylz @Andry-Bal for the info & tips ! Basically, I have changed the version of Java & Maven (I took the same versions as @Andry-Bal). [jamilz@jamil-pc python]$ ./launch.sh getalp.wsd.predict --data_path ../../data/ --weights ../../data/model_weights_wsd0 --disambiguate So, it seems like I had both memory and dependencies compatibility problem. Later after the exams, I might try using @coylz 's advices (training a little model) in order to help future users who will face this problem. |
Hello everyone! :)
I am currently facing issues which are :
I can't launch the evaluate.sh script because of a threading error. I tried my best to find the source of this error, but was unsuccessful.
A screenshot showing this error is made.
About decode.sh and train.sh, the process doesn't stop correctly, which makes the thing not debuggable.
A screenshot showing this error is made.
About prepare.sh, I see a progress bar. Once it is complete, the process doesn't stop, which forces me to press Ctrl+C. Is this normal?
I am currently using my personal computer whose configuration is :
GPU = NVIDIA GeForce GTX960M
CPU = IntelCore i7 6700HQ 2.60GHz
OS = Linux (Manjaro) from Dual Boot
Do you have any suggestion or advice?
Thank you very much for your help.
Best regards,
Jamil
The text was updated successfully, but these errors were encountered: