Explore stable diffusion support in Torch-TRT

[Christina-Young-NVIDIA]

Tasks

Preview Give feedback

1. Exploring Stable Diffusion in Torch Compile Path #2144
   15 of 23
   No Activity Story: Dynamo Compile Improvements component: dynamo +3
   

   
2. 🐛 [Bug] aten.expand fails when rank disagrees with tensor shape #2183
   bug component: converters +2
   
3. feat: Improve Dynamo partitioning System Performance on Large Models #2175
   Story: Dynamo Compile Improvements Story: Export/Compile Unification cla signed component: api [Python] component: build system component: conversion component: dynamo component: lowering component: tests component: torch_compile +10
   
4. fix: Update aten.embedding to reflect schema #2182
   Story: ATen Op Support cla signed component: api [Python] component: conversion component: converters component: dynamo +6
   
5. feat: Add _to_copy, operator.get and clone ATen converters #2161
   Story: Export/Compile Unification cla signed component: api [Python] component: conversion component: converters component: dynamo component: tests +7
   
6. feat: Add Selective ATen decompositions #2173
   Story: Export/Compile Unification cla signed component: api [Python] component: dynamo component: lowering component: torch_compile +6
   
7. Prototype Constant Freezing in Dynamo #2124
   Story: Export/Compile Unification component: dynamo +2
   

[github-actions]

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

[gs-olive]

Full Set of Current Missing Operators for Stable Diffusion

##### Other Sub-Architectures
- torch.ops.aten.bmm.default
- torch.ops.aten.linalg_vector_norm.default
- torch.ops.aten.clamp_min.default
- torch.ops.aten.repeat.default
- torch.ops.aten.full.default
- torch.ops.aten.randn.generator

##### UNet
- torch.ops.aten.arange.start_step + Operator Count: 7
- torch.ops.aten.exp.default + Operator Count: 1
- torch.ops.aten._to_copy.default + Operator Count: 7
- torch.ops.aten.sin.default + Operator Count: 1
- torch.ops.aten.cos.default + Operator Count: 1
- torch.ops.aten.native_group_norm.default + Operator Count: 61
- _operator.getitem + Operator Count: 173
- torch.ops.aten.native_layer_norm.default + Operator Count: 48
- torch.ops.aten._scaled_dot_product_efficient_attention.default + Operator Count: 32
- torch.ops.aten.split.Tensor + Operator Count: 16
- torch.ops.aten.clone.default + Operator Count: 16
- torch.ops.aten._unsafe_index.Tensor + Operator Count: 3