You can deploy this model and start earning money today!

sjmathy/
scare

diffusers

safetensors

Views

No views yet

Vidar Priority27 RoboTwin Training Bundle

This bundle is self-contained for cached-latent LoRA training.

Contents:

code/vidar: Vidar/Wan training code.
data/robotwin_priority27_varcond_cache: portable index.csv, cached latents, cached text embeddings.
models/vidar/vidar.pt: Vidar initialization.
models/Wan2.2-TI2V-5B: base Wan checkpoint. Included: True.
Source videos included for debugging/provenance: False.

Training uses --use_latent_cache, so source videos are not needed for the training dataloader.

Run:

bash

cd /path/to/vidar_priority27_train_bundle_20260426
bash train_priority27_cached_lora.sh /path/to/output

If using Slurm, adapt scripts/sbatch_train_robotwin_priority27_single121_varcond_*.sh to your cluster's partition/GPU flags and point paths inside this bundle.

sjmathy/scare

Vidar Priority27 RoboTwin Training Bundle

sjmathy/
scare