Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add How to Reproduce the Result in README #2

Open
wants to merge 6 commits into
base: master
Choose a base branch
from

Conversation

atnanahidiw
Copy link

@atnanahidiw atnanahidiw commented Sep 26, 2020

Refactor:

  • run_single_task.sh
  • run_all_tasks.sh
  • run_non_pretrained_no_special_token.sh

by:

  • Create scripts/config/model/train.yaml to easily define the model used
  • Make scripts/reproducer.py and Makefile for easy ops

Note: only tested this by printing the command string, sorry 🙏

@gentaiscool gentaiscool self-assigned this Sep 29, 2020
@gentaiscool gentaiscool added the documentation Improvements or additions to documentation label Sep 29, 2020
@atnanahidiw atnanahidiw changed the title Add How to Train in README Add How to Reproduce the Result in README Oct 3, 2020
Copy link
Member

@SamuelCahyawijaya SamuelCahyawijaya left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@gentaiscool : I think we need to change the load_model and clean up the unused model checkpoints that are used for our experiment. We can instead use Indobenchmark model in HF as the replacement. Can you help to clean that part?

@atnanahidiw, later you can follow the checkpoint argument based on the finalized models that we have. So it will be cleaner and easier to follow.

README.md Outdated Show resolved Hide resolved
@gentaiscool
Copy link
Contributor

@gentaiscool : I think we need to change the load_model and clean up the unused model checkpoints that are used for our experiment. We can instead use Indobenchmark model in HF as the replacement. Can you help to clean that part?

@atnanahidiw, later you can follow the checkpoint argument based on the finalized models that we have. So it will be cleaner and easier to follow.

I will clean the code and add the documentation this week including the CONTRIBUTING page. Let's finish the testing and then, we merge this pull request.

@gentaiscool
Copy link
Contributor

@gentaiscool : I think we need to change the load_model and clean up the unused model checkpoints that are used for our experiment. We can instead use Indobenchmark model in HF as the replacement. Can you help to clean that part?
@atnanahidiw, later you can follow the checkpoint argument based on the finalized models that we have. So it will be cleaner and easier to follow.

I will clean the code and add the documentation this week including the CONTRIBUTING page. Let's finish the testing and then, we merge this pull request.

I added the CONTRIBUTING page in the master branch.

@gentaiscool
Copy link
Contributor

@atnanahidiw we just merged the master branch with a new PR #7. Would you mind to check whether this PR has any conflict?

@atnanahidiw
Copy link
Author

hi @gentaiscool, I just resolved all of the conflicts ya, sorry for the late reply 🙏

@atnanahidiw
Copy link
Author

I added the CONTRIBUTING page in the master branch.

and thanks also for the CONTRIBUTING page
should we make it more verbose? ^^a

@gentaiscool
Copy link
Contributor

I added the CONTRIBUTING page in the master branch.

and thanks also for the CONTRIBUTING page
should we make it more verbose? ^^a

Yes, we should, probably in a new PR?

Copy link
Member

@SamuelCahyawijaya SamuelCahyawijaya left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@atnanahidiw : We have some update regarding to the model list, can you help removing the unnecessary models and adding the IndoBERT model. I write the comment for all the models that can be removed and also the list of IndoBERT models.

# lower:
# num_layers:

# # albert-base-uncased-96000
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This model can be removed

# num_layers:
# - 12

# # albert-base-uncased-96000-spm
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This model can be removed

# - 12

# # albert-base-uncased-96000-spm
# - model_checkpoint: albert-base-uncased-96000-spm
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This model can be removed

# num_layers:
# - 12

# # albert-base-uncased-112500-spm
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This model can be removed

# - 12

# scratch
- model_checkpoint: scratch
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This model can be removed

- 24

# babert-bpe-mlm-large-uncased-1100k
- model_checkpoint: babert-bpe-mlm-large-uncased-1100k
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This model can be removed

- 24

# babert-bpe-mlm-uncased-128-dup10-5
- model_checkpoint: babert-bpe-mlm-uncased-128-dup10-5
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This model can be removed

python3 scripts/reproducer.py term-extraction-airy 15 $(BATCH_SIZE) $(HYPERPARAMETER)
python3 scripts/reproducer.py pos-prosa 15 $(BATCH_SIZE) $(HYPERPARAMETER)

reproduce_all_1:
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We can remove reproduce_all_*, it is already covered in reproduce and reproduce_all for this

run_non_pretrained_no_special_token:
python3 scripts/reproducer_non_pretrained.py $(DATASET) $(EARLY_STOP) $(BATCH_SIZE)

run_non_pretrained_no_special_token_all:
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There are 8 tasks in here, can you please help adding the other 4 similar to the list in the reproduce_all?

- model_checkpoint: babert-bpe-mlm-uncased-128-dup10-5
lower: True
num_layers:
- 12
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you help adding 8 indoBERT models in this file, the model checkpoint and the num_layers would be as follow:

  • indobenchmark/indobert-base-p1 | 12 layers
  • indobenchmark/indobert-base-p2 | 12 layers
  • indobenchmark/indobert-large-p1 | 24 layers
  • indobenchmark/indobert-large-p2 | 24 layers
  • indobenchmark/indobert-lite-base-p1 | 12 layers
  • indobenchmark/indobert-lite-base-p2 | 12 layers
  • indobenchmark/indobert-lite-large-p1 | 24 layers
  • indobenchmark/indobert-lite-large-p2 | 24 layers

@SamuelCahyawijaya
Copy link
Member

I added the CONTRIBUTING page in the master branch.

and thanks also for the CONTRIBUTING page
should we make it more verbose? ^^a

Yes, we should, probably in a new PR?

Yeah I agree, this one can be a new PR. Thank you 😀

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
documentation Improvements or additions to documentation
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants