Fix integration tests for SAM

[qubvel]

What does this PR do?

Fix device placement for tensors for pytorch integration tests of SAM model

[qubvel]

run-slow: sam

[github-actions]

This comment contains run-slow, running the specified jobs: This comment contains run-slow, running the specified jobs:

models: ['models/sam']
quantizations: [] ...

[qubvel]

Pytorch tests are fixed; only TF fails left, won't fix

FAILED tests/models/sam/test_modeling_tf_sam.py::TFSamModelIntegrationTest::test_inference_mask_generation_three_boxes_point_batch - TypeError: Cannot convert '__empty__' to EagerTensor of dtype int32
FAILED tests/models/sam/test_modeling_tf_sam.py::TFSamModelIntegrationTest::test_inference_mask_generation_two_points_point_batch - TypeError: Cannot convert '__empty__' to EagerTensor of dtype int32
FAILED tests/models/sam/test_processor_sam.py::TFSamProcessorTest::test_rle_encoding - TypeError: Only integers, slices (`:`), ellipsis (`...`), tf.newaxis (`None`) and scalar tf.int32/tf.int64 tensors are valid indices, got <tf.Tensor: shape=(0,), dtype=bool, numpy=array([], dtype=bool)>
===== 3 failed, 109 passed, 192 skipped, 26 warnings in 217.18s (0:03:37) ======

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[ydshieh]

Thank you @qubvel for your kindness!

Hope one day torch.testing.assert_close can take care of this on its own ...