fix tests/models/edgetam_video/test_modeling_edgetam_video.py

[sywangyi]

and tests/models/sam2_video/test_modeling_sam2_video.py crash

fix crash like
FAILED tests/models/edgetam_video/test_modeling_edgetam_video.py::EdgeTamVideoModelIntegrationTest::test_inference_mask_generation_video_multi_objects_multi_points - AttributeError: positional_embedding is not an nn.Parameter

[github-actions]

[For maintainers] Suggested jobs to run (before merge)

run-slow: edgetam_video, sam2_video

[sywangyi]

@SunMarc please help review.
wo the fix the tests/models/sam2_video/test_modeling_sam2_video.py and tests/models/edgetam_video/test_modeling_edgetam_video.py all fail

[SunMarc]

Thanks for checking, while this might fix the tests, I'm sure this will probably lead to another issues as . cc @molbap @ArthurZucker

The comment below is probably connected to this issue:

# need to be ignored, as it's a buffer and will not be correctly detected as tied weight
_keys_to_ignore_on_load_missing = ["prompt_encoder.shared_embedding.positional_embedding"]

[sywangyi]

Thanks for checking, while this might fix the tests, I'm sure this will probably lead to another issues as . cc @molbap @ArthurZucker

The comment below is probably connected to this issue:

# need to be ignored, as it's a buffer and will not be correctly detected as tied weight
_keys_to_ignore_on_load_missing = ["prompt_encoder.shared_embedding.positional_embedding"]

I see https://github.com/huggingface/transformers/blob/64c12fdf5f6a35704f2932db533c76973b1e9a30/src/transformers/models/sam_hq/modular_sam_hq.py#L445， it only contains _keys_to_ignore_on_load_missing wo tied weight

[sywangyi]

same with #42840, so could be closed

[SunMarc]

Thanks !