Skip to content

Question about training time and compute resources for reproducing Spatial-TTT #6

Description

@WangHanLinHenry

Thank you for releasing the code and the Spatial-TTT-Data-97k dataset. I am trying to reproduce the training using the provided spatial_ttt_train.sh script.

I noticed that the README mentions training with Spatial-TTT-Data-97k, lact_chunk_size=2648, window_size=2648, video_max_frames=128, and 8 GPUs by default. However, I could not find detailed information about the expected training time and compute requirements.

Could you please share some details about the training resources used for the released/reproduction model?

Specifically, I would like to ask:

  1. What GPU type and number of GPUs were used for training?
  2. Approximately how long does training on Spatial-TTT-Data-97k take, in wall-clock time or GPU hours?

Activity

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