Assert seq_lens.max() <= seq_len

[onoregleb]

I have a script:

torchrun --nproc_per_node=4 generate.py \
  --task i2v-14B \
  --size 480*832 \
  --ckpt_dir /mnt/tank/scratch/edubskiy/Wan2.1/I2V-14B-480P \
  --image img.jpg \
  --frame_num 33 \
  --t5_cpu \
  --offload_model True \
  --sample_steps 40 \
  --save_file output_video.mp4 \
  --dit_fsdp --t5_fsdp --ulysses_size 4 \
  --prompt "A young man enjoying a relaxed photoshoot in a sunlit studio. He is casually moving, adjusting his posture with ease, and smiling naturally. The photographer in the background captures candid moments as he playfully runs a hand through his hair. The setting is modern and minimalistic, with soft natural light casting gentle shadows on the white backdrop."

When I run it, the error occurs: assert seq_lens.max() <= seq_len

This occurs when I change the frame_nums parameter, even though it follows the 4n+1 rule. How can I fix this?