Regression in SAM2 caused by changes in mark_tied_weights_as_initialized #42837
Description
System Info
transformersversion: 5.0.0.dev0 (commit hash: 37426b2)- Platform: Linux-6.6.87.2-microsoft-standard-WSL2-x86_64-with-glibc2.35
- Python version: 3.11.0rc1
- Huggingface_hub version: 1.2.2
- Safetensors version: 0.7.0
- Accelerate version: 1.12.0
- Accelerate config: not found
- DeepSpeed version: not installed
- PyTorch version (accelerator?): 2.8.0+cu128 (CUDA)
- Using distributed or parallel set-up in script?: No
- Using GPU in script?: Yes
- GPU type: NVIDIA GeForce RTX 4080
Who can help?
Information
- The official example scripts
- My own modified scripts
Tasks
- An officially supported task in the
examplesfolder (such as GLUE/SQuAD, ...) - My own task or dataset (give details below)
Reproduction
Script to reproduce:
from transformers import (
Sam2VideoModel,
)
model = Sam2VideoModel.from_pretrained(
"facebook/sam2.1-hiera-large",
revision="665f8e2ad61cf5f53d65644ff27c8ee525124610",
)
Exception:
File "/home/krwaraksa/code/SpatialAI2/repro_sam2.py", line 5, in <module>
model = Sam2VideoModel.from_pretrained(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/krwaraksa/code/SpatialAI2/sam3_venv/lib/python3.11/site-packages/transformers/modeling_utils.py", line 248, in _wrapper
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/home/krwaraksa/code/SpatialAI2/sam3_venv/lib/python3.11/site-packages/transformers/modeling_utils.py", line 3988, in from_pretrained
model, missing_keys, unexpected_keys, mismatched_keys, offload_index, error_msgs = cls._load_pretrained_model(
^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/krwaraksa/code/SpatialAI2/sam3_venv/lib/python3.11/site-packages/transformers/modeling_utils.py", line 4148, in _load_pretrained_model
model.mark_tied_weights_as_initialized()
File "/home/krwaraksa/code/SpatialAI2/sam3_venv/lib/python3.11/site-packages/transformers/modeling_utils.py", line 4476, in mark_tied_weights_as_initialized
param = self.get_parameter(tied_param)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/krwaraksa/code/SpatialAI2/sam3_venv/lib/python3.11/site-packages/torch/nn/modules/module.py", line 847, in get_parameter
raise AttributeError("`" + param_name + "` is not an nn.Parameter")
AttributeError: `positional_embedding` is not an nn.Parameter
This is likely caused by this change: #42270, in mark_tied_weights_as_initialized(). Before AttributeError was silently ignored, now it's getting thrown.
Expected behavior
Model gets loaded. It works in transformers 4.57.3