Release 2.1.1 - HunyuanVideo

Updates

• Introduces HunyuanVideo to validator challenges
• Upgraded to diffusers 0.32.2
• Added random generation parameter selection for resolution, number of inference steps and number of frames. See bitmind/validator/config.py for configured minimums, maximums, and options for each model.
• Model-specific FPS settings for saving video outputs
• Removes score decay for miners unselected uids
• Increased video rewards to 40%

Validator Update Process

Important: If you have less than the recommended 600GB storage, you may run out of storage with this upgrade. New recommended storage is 1TB to provide buffer for future releases, rather than requesting incremental upgrades.

• Autoupdate Validators: No action necessary.
• Self-managed Validators: Run ./setup_env.sh and restart your validator.
  • The setup script will update your diffusers version.
  • The HunyuanVideo model added in this release will be downloaded when the validator starts back up

Current Validator Storage Overview

• Total storage used by SN34 is < 450GB including repository, models, cached data and conda environment
• HunyuanVideotakes up an additional 16GB of space in the Huggingface cache
• ~/.cache/huggingface (generative models) is now 221GB.
• ~/.cache/sn34(real and synthetic data) is ~200GB (unchanged)