Problem about hunyuan worldplay pipeline

[hzeroyuke]

When running HunyuanWorldPlayPipeline.from_pretrained(...), the pipeline fails to load the ByT5 glyph processor with the following error:

RuntimeError: Error loading byT5 glyph processor

The root cause is a hard version check introduced in recent versions of transformers(4.57,3), which now requires PyTorch >= 2.6 (But docs/install.md requires torch==2.5.1 for hunyuan worldplay)to load .pt checkpoint files via torch.load, due to a serious security vulnerability

Traceback

File "hunyuan_worldplay_synthesis.py", line 397, in _load_byt5
    byt5_kwargs = load_glyph_byT5_v2(byt5_args, device=device)
  File "byT5/__init__.py", line 150, in load_byt5_and_byt5_tokenizer
    byt5_text_encoder = T5ForConditionalGeneration.from_pretrained(
  File "transformers/utils/import_utils.py", line 1647, in check_torch_load_is_safe
    raise ValueError(
ValueError: Due to a serious vulnerability issue in `torch.load`, even with
`weights_only=True`, we now require users to upgrade torch to at least v2.6
in order to use the function. This version restriction does not apply when
loading files with safetensors.
See: https://nvd.nist.gov/vuln/detail/CVE-2025-32434

Root Cause

checkpoints/HunyuanVideo-1.5/text_encoder/Glyph-SDXL-v2/checkpoints/byt5_model.pt is a legacy .pt (pickle) format file. The transformers library now blocks torch.load on .pt files for PyTorch < 2.6 as a security measure. The restriction does not apply to .safetensors format.

Steps to Reproduce

CUDA_VISIBLE_DEVICES=0 python test/test_hunyuan_worldplay.py

[zengbohan0217]

Thank you very much for pointing out this issue. We will have our colleagues improve the environment setup script for installing dependencies.