dataset & toml
#1
by xilai0715 - opened
This model is amazing. Could you please open-source your dataset and the trained toml file? I would like to train the model myself. Looking forward to your reply. Hope I didn't bother you.
I want to re-caption it with my latest script, which uses gemma3 27b for stills and narrative, and looks at 24 vs 12 frames. Once I do that, I will upload the dataset. Thank you.
config:
# general configurations
[general]
resolution = [640, 360]
caption_extension = ".txt"
batch_size = 1
enable_bucket = true
bucket_no_upscale = false
[[datasets]]
video_directory = "/home/anon/Documents/musubi-tuner/data/city-video"
cache_directory = "/home/anon/Documents/musubi-tuner/data/city-video-cache/"
target_frames = [1, 5, 9, 13, 17, 21, 25, 29, 33, 37, 41, 45, 49, 53, 57, 61, 65, 69, 73, 77, 81]
frame_extraction = "uniform"
source_fps = 23.98
I had to scale the video files down to 640x360 or I got OOM errors with 48GB VRAM.
Here is the dataset: https://huggingface.co/datasets/quarterturn/cta-852x480-23.98-324
Thank you very much. Looking forward to your new model.