Bo Wang

fix: fix the "schema not found for node" error

2

Signed-off-by: Bo Wang # Description In partitioning, when we check if a node would modify the input value, we use the node->schema() function from PyTorch previously. However, this function could...

Signed-off-by: Bo Wang # Description Before this feature https://github.com/pytorch/TensorRT/issues/1031 is supported, there might be some cases that TensorRT segmented block has NonTensor input, so we go through the nodes' output...

feat: keep model's original input signature

4

Signed-off-by: Bo Wang # Description There are some requests that users doesn't want to change the models' inputs signature for input, which could make it difficult to use in a...

**Is your feature request related to a problem? Please describe.** #1024 There are some models that LowerGraph() might introduce new input to the entire graph after lowering, which may cause...

## Bug Description We are getting this kind of ivalue type https://github.com/pytorch/pytorch/blob/b34b192d6b97325c9f78e5995c48c8498ede34bd/aten/src/ATen/core/ivalue.h#L159, however, when it's used by a later subgraph in fallback shape analysis, there is no available types that...

**Is your feature request related to a problem? Please describe.** We have this graph: ``` INFO: [Torch-TensorRT] - Segment Block @171: Target: Torch Graph: graph(%1 : bool, %prev_features.1 : Tensor,...

feat: rewriting param to a Constant if it's a introduced input

1

Signed-off-by: Bo Wang # Description In some cases, some parameters are graphs' input. We can rewrite them into constants to make sure it works. Fixes #1190 ## Type of change...

✨[Feature] Support auto data type transformation between int64 <-> int32

2

**Is your feature request related to a problem? Please describe.** In this graph: ``` INFO: [Torch-TensorRT - Debug Build] - Partitioned Graph: [Segment Block @0: Target: TensorRT Graph: graph(%index.1 :...

## Bug Description In this graph: ``` %1: Int[] = prim::ListConstruct(%w, %h) prim::If (%0): block0(): %3 = str = aten::format(%3, %4, %1) block1(): ... ``` In partitioning stage, prim::If is...

**Is your feature request related to a problem? Please describe.** Currently Torch-TensorRT is relying on Partitioning to support Tuple/List grouped inputs, which means for some models with Tuple/List input/output, they...