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

Does this project only work on linux? #47

Open
xhz2003 opened this issue Oct 9, 2024 · 6 comments
Open

Does this project only work on linux? #47

xhz2003 opened this issue Oct 9, 2024 · 6 comments

Comments

@xhz2003
Copy link

xhz2003 commented Oct 9, 2024

How to run this biomedgpt big model locally? Any recommended links to related tutorials?

@taokz
Copy link
Owner

taokz commented Oct 16, 2024

Currently, the Fairseq version is complete but works only on Linux due to limitations with Fairseq itself. However, for inference, you can refer to the Colab example in the README, which uses the transformers library and can be executed on other operating systems.

In addition, could you clarify which tutorials you’re looking for? Are you asking for a tutorial on Fairseq?

@xhz2003
Copy link
Author

xhz2003 commented Oct 22, 2024

Currently, the Fairseq version is complete but works only on Linux due to limitations with Fairseq itself. However, for inference, you can refer to the Colab example in the README, which uses the transformers library and can be executed on other operating systems.

In addition, could you clarify which tutorials you’re looking for? Are you asking for a tutorial on Fairseq?

Thank you.
1, why after opening the train.py file inside pycharm:
from fairseq import (
# checkpoint_utils,
options, quantization_utils,
quantization_utils,
tasks,
utils.
)
Is the fairseq you mentioned a file directory? Why are all the references in parentheses missing? (That is, I looked at all the files you gave me and found that there are no files (options, quantization_utils,tasks,utils,) under the fairseq file.
2、What file is the main network structure of the BiomedGPT model you gave me?
3、Can I modify the network structure of BiomedGPT? And then fine-tune it so that it handles other related subtasks. Do you have any better suggestions?

@xhz2003
Copy link
Author

xhz2003 commented Oct 22, 2024

Currently, the Fairseq version is complete but works only on Linux due to limitations with Fairseq itself. However, for inference, you can refer to the Colab example in the README, which uses the transformers library and can be executed on other operating systems.

In addition, could you clarify which tutorials you’re looking for? Are you asking for a tutorial on Fairseq?

May I ask, in which file path are your trained models saved? I want to fine tune him.

@taokz
Copy link
Owner

taokz commented Oct 24, 2024

  1. There are separate files for tasks, utils, models, etc. I’m not sure why PyCharm is reporting errors—it could be due to an incorrectly configured environment.

  2. The model definition can be found in models/ofa/ofa.py.

  3. Yes, feel free to modify the structure. If you’d like to add new tasks, you can create a new dataset and task by adding Python files to the data and tasks directories.

  4. The checkpoints are stored on both HuggingFace and Dropbox. Please refer to the README file and follow the steps outlined there.

@xhz2003
Copy link
Author

xhz2003 commented Oct 27, 2024

In the path of models/ofa you mentioned, the directories at the same level as it include clip, taming, as well as init.py, search.py, sequence_gemerator.py, what is the connection and difference between them? Can you briefly describe the relationship between the various files throughout the project? Which ones need core attention?

@taokz
Copy link
Owner

taokz commented Oct 30, 2024

The sequence_generator.py script is used to decode or generate outputs based on specified hyperparameters, such as model selection and search strategies (e.g., beam search defined in search.py). The taming module, inherited from VQGAN (https://github.com/CompVis/taming-transformers), is used to encode image patches for preprocessing tasks (refer to masked image modeling in the paper) or to generate image patches. The clip module defines image encoders, such as ResNet and ViT, for training.

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

No branches or pull requests

2 participants