↔ [Converter] Add support for assorted operators in the FX aten path

[gs-olive]

aten.unsqueeze, aten.reshape, aten.permute, aten.transpose

• Function Schema:

  • torch.ops.aten.arange.start: ((), {})
  • torch.ops.aten.rsub.Scalar: ((torch.float32,), {})
  • torch.ops.aten._to_copy.default: ((torch.int32,), {})
  • torch.ops.aten.embedding.default: ((torch.float32, torch.int64), {})
  • torch.ops.aten.embedding.default: ((torch.float32, torch.int32), {})
  • torch.ops.aten.layer_norm.default: ((torch.float32, None, torch.float32, torch.float32), {})
  • torch.ops.aten.addmm.default: ((torch.float32, torch.float32, torch.float32), {})
  • torch.ops.aten._softmax.default: ((torch.float32,), {})
  • torch.ops.aten.where.self: ((torch.bool, torch.float32, torch.float32), {})

• Original PyTorch API: torch.arange, torch.embedding, torch.layer_norm, torch.addmm, torch._softmax, torch.where

• Relevant TensorRT Documentation: IElementWiseLayer, IConstantLayer

Add support for the above function schemas as aten converters.

[apbose]

Added #1790 , #1794, #1796, #1797, #1798, #1799 for the subtasks

[narendasan]

#1724 (additional sub tasks)

[github-actions]

This issue has not seen activity for 90 days, Remove stale label or comment or this will be closed in 10 days