Loading a checkpoint with tied weights results in `ValueError: This checkpoint seem corrupted.`

[tomaarsen]

System Info

• transformers version: 5.0.0.dev0
• Platform: Windows-10-10.0.26100-SP0
• Python version: 3.11.13
• Huggingface_hub version: 1.1.5
• Safetensors version: 0.6.2
• Accelerate version: 1.11.0
• Accelerate config: not found
• DeepSpeed version: not installed
• PyTorch version (accelerator?): 2.9.0+cu126 (CUDA)
• Using distributed or parallel set-up in script?: No
• Using GPU in script?: No
• GPU type: NVIDIA GeForce RTX 3090

Who can help?

@Cyrilvallez @molbap

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction

from transformers import AutoModel

model_name = "sentence-transformers/sentence-t5-base"
model = AutoModel.from_pretrained("sentence-transformers/sentence-t5-base")
print(type(model))

Currently, the behaviour on main is:

Traceback (most recent call last):
  File "c:\code\transformers\demo_sentence_t5.py", line 5, in <module>
    model = AutoModel.from_pretrained("sentence-transformers/sentence-t5-base")
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\code\transformers\src\transformers\models\auto\auto_factory.py", line 373, in from_pretrained
    return model_class.from_pretrained(
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\code\transformers\src\transformers\modeling_utils.py", line 275, in _wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "C:\code\transformers\src\transformers\modeling_utils.py", line 3977, in from_pretrained
    model, missing_keys, unexpected_keys, mismatched_keys, offload_index, error_msgs = cls._load_pretrained_model(
                                                                                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "C:\code\transformers\src\transformers\modeling_utils.py", line 4163, in _load_pretrained_model
    model.tie_weights(missing_keys=missing_keys, recompute_mapping=False)
  File "C:\code\transformers\src\transformers\modeling_utils.py", line 2337, in tie_weights
    raise ValueError(
ValueError: This checkpoint seem corrupted. The tied weights mapping for this model specifies to tie shared.weight (which should be present and is not), to encoder.embed_tokens.weight (which is present).

Expected behavior

I expect the model to load normally, as it had been doing for <v5.0.

• Tom Aarsen

[vasqu]

Hey @tomaarsen 👋

this is expected for now but @Cyrilvallez wanted to add a PR to fix this. The core issue is that while the tied mapping says x should exist, it's the other way around.

Related PRs #42377, #42355

[vasqu]

Should be fixed by #42507