8000 How to train inf-dit · Issue #28 · zai-org/Inf-DiT · GitHub
[go: up one dir, main page]

Skip to content

How to train inf-dit #28

Open
Open

Description

@billfjj

I want to fine tune this model, but i am unable to run your training code. Could you please provide a detailed description of the dataset fromat used for your training?

Activity

  1. ivcylc commented on Nov 2, 2024

    @ivcylc

    Have you succeeded in training it?
    I am wondering where can i change the batch_size

  2. yzy-thu commented on Nov 2, 2024

    @yzy-thu
    Contributor

    I want to fine tune this model, but i am unable to run your training code. Could you please provide a detailed description of the dataset fromat used for your training?

    webdataset with .jpg

  3. yzy-thu commented on Nov 2, 2024

    @yzy-thu
    Contributor

    Have you succeeded in training it? I am wondering where can i change the batch_size

    scripts/ds_config_zero.json

  4. ivcylc commented on Nov 2, 2024

    @ivcylc

    Have you succeeded in training it? I am wondering where can i change the batch_size

    scripts/ds_config_zero.json

    Thanks, and i found the loss when training with bs=2 is large than that with bs=1
    Additionally, insterestly, when bs=2, the predicted seems excatedly half of that when bs=1, which cause error

  5. billfjj commented on Nov 4, 2024

    @billfjj
    Author

    I want to fine tune this model, but i am unable to run your training code. Could you please provide a detailed description of the dataset fromat used for your training?

    webdataset with .jpg

    I built a tar file with only jpg, but after changing the path of this tar file to the configuration file, the training will report an error that the dataset class does not have the len attribute.

  6. BrooksXiaoxi commented on Nov 12, 2024

    @BrooksXiaoxi

    Have you succeeded in training it? I am wondering where can i change the batch_size

    scripts/ds_config_zero.json

    Thanks, and i found the loss when training with bs=2 is large than that with bs=1 Additionally, insterestly, when bs=2, the predicted seems excatedly half of that when bs=1, which cause error

    Could you please tell me how you prepared your training data and how you configured the model to train? I am having trouble fine-tuning my own model on my dataset. I would appreciate your response, thanks!!!

  7. sunrccc commented on Dec 6, 2024

    @sunrccc

    Have you succeeded in training it? I am wondering where can i change the batch_size

    scripts/ds_config_zero.json

    Thanks, and i found the loss when training with bs=2 is large than that with bs=1 Additionally, insterestly, when bs=2, the predicted seems excatedly half of that when bs=1, which cause error

    Could you please tell me how you prepared your training data and how you configured the model to train? I am having trouble fine-tuning my own model on my dataset. I would appreciate your response, thanks!!!

    May I ask if you have solved this problem? Could you give me some details?

  8. sunrccc commented on Dec 6, 2024

    @sunrccc

    May I ask if you have solved this problem? Could you give me some details?

    May I ask if you have solved this problem? Could you give me some details?

  9. sunrccc commented on Dec 6, 2024

    @sunrccc

    I want to fine tune this model, but i am unable to run your training code. Could you please provide a detailed description of the dataset fromat used for your training?

    webdataset with .jpg

    Sorry, I don't quite understand your answer. Could you explain it in detail? Thank you very much.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions

      0