Pytorch Sweeps tracing - ttnn.subtract - Low PCC and other fails

[VirdhatchaniKN]

Tracking Issue : #13521
File : tests/sweep_framework/sweeps/eltwise/binary/subtract/subtract_tensor_fails.py

Issues faced :

• round_up: multiple must not be 0 when one of the tensor dimension size is 0 - PASSING
• PCC Drop due to broadcasting

Details on above listed issues : [ Testing Branch : virdhatchani/sub_pcc_sweeps ]

• PCC Drop due to broadcasting - Analysis_of_subtract.pdf

  • Shapes :
    • {'shape': [1, 10], 'other': [10, 1]}
    • {'shape': [1, 15], 'other': [15, 1]}
    • {'shape': [1, 17], 'other': [17, 1]}
    • {'shape': [1, 2], 'other': [2, 1]}
    • {'shape': [16, 1, 49], 'other': [16, 49, 1]}
    • {'shape': [16, 1, 64], 'other': [16, 64, 1]}
    • {'shape': [24, 1], 'other': [1, 24]}
    • {'shape': [4, 1, 49], 'other': [4, 49, 1]}
    • {'shape': [4, 1, 64], 'other': [4, 64, 1]}
    • {'shape': [64, 1, 49], 'other': [64, 49, 1]}
    • {'shape': [64, 1, 64], 'other': [64, 64, 1]}

• round_up: multiple must not be 0 - Case 3 in Support for creating a 0-dim tensor when a single scalar value or shape [ ] is given #13803

  • Shapes :
    • {"shape": [0, 1], "other": [0, 1]} - #0: Add support for 0-volume and 1-volume tensors for ttnn::add #14611 : Fix merged to main, Shape passing
    • {"shape": [0], "other": [0]} - Passes with temp fix from #13536: testing without 0-dim and 1D restrictions #14280