Provided weights are incompatible with the code

The repo claims to implement a document image parsing model (DonutModel), but the provided weights (dolphin_model.bin, Hugging Face model.safetensors) are from a text-only LLM and cannot be loaded.

demo_page.py expects vision encoder weights.

load_state_dict() throws shape mismatch and missing key errors.

No compatible dolphin-2.0-base.pt is provided.

Please clarify: where are the correct weights for the visual document parsing model, or is this repo incomplete?

Inspect demo_page.py and compare its expected vision encoder weights with dolphin_model.bin and the Hugging Face model.safetensors files. Reproduce the load_state_dict() shape-mismatch and missing-key errors, then determine whether compatible dolphin-2.0-base.pt weights exist or whether the repository is incomplete.

Written by the indexing model from the issue text.