Closed AradAshrafi closed 4 years ago
Not sure if this is specific to colab. It looks like you haven't set the data_dir variable?
export data_dir=</path/to/data_dir>
Not sure if this is specific to colab. It looks like you haven't set the data_dir variable?
export data_dir=</path/to/data_dir>
Thanks for your reply. Yes, I include the codes to export data_dir, but it seems that Colab environment doesn't export the variable. I added a script file that includes export and python train.py, and it worked! Also, I had troubles in building .jsonlines files in setup_training process. I built these files in my system locally, then I uploaded them to my drive and used them.
Hie @AradAshrafi ,
Can you please share your code snippet of how you resolved it by including in the script itself..??
Hi @Atul-Anand-Jha I am sorry for my delay. I resolve the exporting issue by running the related commands and the export itself inside a bash script file like:
#!/bin/bash
export data_dir="conll-2012"
CUDA_VISIBLE_DEVICES=0 python3 train.py train_spanbert_base
Thanks for the reply. Btw, I also figured out a way of doing this via Python script itself; using os.envion[""]
setup function.
Hi again @Atul-Anand-Jha, Have you changed the .conf file because of the recent changes in the Colab's environment? It killed the process because of the lack of GPU memory. Also, Have you tried the code on other languages?
Hi again @Atul-Anand-Jha, Have you changed the basic.conf file because of the recent changes in the Colab's environment? It killed the process because of the lack of GPU memory. Also, Have you tried the code on other languages?
No, I tried Python language only, no other languages to implement this.
And, I didn't change any configuration file as well. I did few small changes but in the main - code itself, not in the .conf.
What kind of exact bug you are getting ?
Hi again @Atul-Anand-Jha, Have you changed the basic.conf file because of the recent changes in the Colab's environment? It killed the process because of the lack of GPU memory. Also, Have you tried the code on other languages?
No, I tried Python language only, no other languages to implement this.
And, I didn't change any configuration file as well. I did few small changes but in the main - code itself, not in the .conf.
What kind of exact bug you are getting ?
Actually, It is not a bug. I get out of memory error because of the Colab's recent limitations. I was using the spanbert_base mode, and I needed to change some parameters to make this out of memory error disappear (e.g. lowering the ffnn_size). I was looking for an optimized configuration. Have you tried the model on other datasets?
I'm assuming this is solved and closing this. Feel free to re-open if you need to. Also leaving a link to Jonathan's Colab notebook in case someone comes across this thread later.
谢谢回复。顺便说一句,我还想出了一种通过Python脚本本身完成此操作的方法。使用
os.envion[""]
设置功能。
Thank you so much!
Is here anyone using google colab to train this model? I have faced many problems such as: 1- The model was unable to create gold_conll files in colab, I have built them in my system locally and then uploaded it to my drive to solve this. 2- I am unable to train the model, it reports a problem related to pyhocon such as follows:
any help with this?