#8364: Enable enable_fast_runtime_mode by default

[ayerofieiev-tt]

Currently, enable_fast_runtime_mode is set to False in ttnn but we want the users to run at the full speed by default

• Deleted allow_to_fallback_to_golden_function_on_failure flag
• Skip tests which fallback to torch
• Updated model tests to manually fallback to torch in the model/test code
• Set some unit tests to run with enable_fast_runtime_mode=False to test features like validation, logging, etc
• Set enable_fast_runtime_mode to True in ttnn

Test Changes:

Changes to test_concat.py | Test passing: 2/11

Tests skipped in test_concat as ttnn.concat requires Padded Tiles:

• [dim=0-width=4-height=20]
• [dim=0-width=4-height=32]
• [dim=0-width=32-height=20]
• [dim=1-width=4-height=20]
• [dim=1-width=4-height=32]
• [dim=1-width=32-height=20]

Tests skipped in test_sharded_concat as ttnn.concat only supports Layout.TILE_LAYOUT:

• [input_shape_a=(1, 1, 16, 16)-shard_shape_a=(8, 16)-input_shape_b=(1, 1, 16, 16)-shard_shape_b=(8, 16)-output_shard_shape=(8, 32)-shard_grid={[(x=0,y=0) - (x=0,y=1)]}]
• [input_shape_a=(1, 1, 160, 32)-shard_shape_a=(80, 32)-input_shape_b=(1, 1, 160, 32)-shard_shape_b=(80, 32)-output_shard_shape=(80, 64)-shard_grid={[(x=0,y=0) - (x=0,y=1)]}]
• [input_shape_a=(1, 1, 160, 32)-shard_shape_a=(80, 32)-input_shape_b=(1, 1, 160, 16)-shard_shape_b=(80, 16)-output_shard_shape=(80, 48)-shard_grid={[(x=0,y=0) - (x=0,y=1)]}]

test_repeat_interleave.py (0/2)

Tests skipped as ttnn.repeat_interleave does not support non-4D tensors:

• test_repeat_interleave
• test_repeat_interleave_with_repeat_tensor

Changes to test_split.py | Tests passing: 0/4

Tests skipped asttnn.split is not implemented:**

• [dim=-1-split_size=2-w=64-h=32]
• [dim=-1-split_size=4-w=64-h=32]
• [dim=-2-split_size=2-w=64-h=32]
• [dim=-2-split_size=4-w=64-h=32]

Changes to test_getitem.py | Tests passing: 32/57

Tests skipped as Tensor.__getitem__ only supports tensors stored on device:

• test_getitem
  • [on_device=False-input_layout=Layout.ROW_MAJOR-width=32-height=32-batch_sizes=()]
  • [on_device=False-input_layout=Layout.ROW_MAJOR-width=32-height=32-batch_sizes=(1,)]
  • [on_device=False-input_layout=Layout.ROW_MAJOR-width=32-height=64-batch_sizes=()]
  • [on_device=False-input_layout=Layout.ROW_MAJOR-width=32-height=64-batch_sizes=(1,)]
  • [on_device=False-input_layout=Layout.ROW_MAJOR-width=96-height=32-batch_sizes=()]
  • [on_device=False-input_layout=Layout.ROW_MAJOR-width=96-height=32-batch_sizes=(1,)]
  • [on_device=False-input_layout=Layout.ROW_MAJOR-width=96-height=64-batch_sizes=()]
  • [on_device=False-input_layout=Layout.ROW_MAJOR-width=96-height=64-batch_sizes=(1,)]
  • [on_device=False-input_layout=Layout.TILE-width=32-height=32-batch_sizes=()]
  • [on_device=False-input_layout=Layout.TILE-width=32-height=32-batch_sizes=(1,)]
  • [on_device=False-input_layout=Layout.TILE-width=32-height=64-batch_sizes=()]
  • [on_device=False-input_layout=Layout.TILE-width=32-height=64-batch_sizes=(1,)]
  • [on_device=False-input_layout=Layout.TILE-width=96-height=32-batch_sizes=()]
  • [on_device=False-input_layout=Layout.TILE-width=96-height=32-batch_sizes=(1,)]
  • [on_device=False-input_layout=Layout.TILE-width=96-height=64-batch_sizes=()]
  • [on_device=False-input_layout=Layout.TILE-width=96-height=64-batch_sizes=(1,)]
• test_getitem_2d
  • [on_device=False-input_layout=Layout.ROW_MAJOR-width=32-height=32]
  • [on_device=False-input_layout=Layout.ROW_MAJOR-width=32-height=64]
  • [on_device=False-input_layout=Layout.ROW_MAJOR-width=96-height=32]
  • [on_device=False-input_layout=Layout.ROW_MAJOR-width=96-height=64]
  • [on_device=False-input_layout=Layout.TILE-width=32-height=32]
  • [on_device=False-input_layout=Layout.TILE-width=32-height=64]
  • [on_device=False-input_layout=Layout.TILE-width=96-height=32]
  • [on_device=False-input_layout=Layout.TILE-width=96-height=64]
• test_getitem_scalar_output (AssertionError: assert ttnn.Tensor.getitem: cannot return a scalar! in "")