Bump TensorRT visible torch version

[Isotr0py]

Related issue: #69

Though torch2.0 will raise UnsupportedOperatorError while exporting onnx, the nightly torch2.1 can use TensorRT normally. It seems that torch has fixed the issue in their coming version.

TensorRT test with torch2.1 nightly:

TensorRT version: 8.6.1

...

======= Diagnostic Run torch.onnx.export version 2.1.0.dev20230511+cu118 =======
verbose: False, log level: Level.ERROR
======================= 0 NONE 0 NOTE 0 WARNING 0 ERROR ========================

[I] Folding Constants | Pass 1
[I] Inferring shapes in the model with `onnxruntime.tools.symbolic_shape_infer`.
    Note: To force Polygraphy to use `onnx.shape_inference` instead, set `allow_onnxruntime=False` or use the `--no-onnxruntime-shape-inference` command-line option.
[I]     Total Nodes | Original:  4544, After Folding:  3269 |  1275 Nodes Folded
[I] Folding Constants | Pass 2
[I]     Total Nodes | Original:  3269, After Folding:  2530 |   739 Nodes Folded
[I] Folding Constants | Pass 3
[I]     Total Nodes | Original:  2530, After Folding:  2530 |     0 Nodes Folded
Building TensorRT engine for models/accelerate/tensorrt/gsdf/Counterfeit-V2.5/onnx/unet.opt.onnx: models/accelerate/tensorrt/gsdf/Counterfeit-V2.5/engine/unet.plan
[W] onnx2trt_utils.cpp:374: Your ONNX model has been generated with INT64 weights, while TensorRT does not natively support INT64. Attempting to cast down to INT32.
[W] onnx2trt_utils.cpp:400: One or more weights outside the range of INT32 was clamped
[I]     Configuring with profiles: [Profile().add('sample', min=(2, 4, 32, 32), opt=(2, 4, 64, 64), max=(2, 4, 128, 128)).add('encoder_hidden_states', min=(2, 77, 768), opt=(2, 77, 768), max=(2, 77, 768)).add('timestep', min=[1], opt=[1], max=[1])]
[I] Building engine with configuration:
    Flags                  | [FP16]
    Engine Capability      | EngineCapability.DEFAULT
    Memory Pools           | [WORKSPACE: 10902.81 MiB, TACTIC_DRAM: 15101.81 MiB]
    Tactic Sources         | []
    Profiling Verbosity    | ProfilingVerbosity.DETAILED
    Preview Features       | [FASTER_DYNAMIC_SHAPES_0805, DISABLE_EXTERNAL_TACTIC_SOURCES_FOR_CORE_0805]
[W] TensorRT encountered issues when converting weights between types and that could affect accuracy.
[W] If this is not the desired behavior, please modify the weights or retrain with regularization to adjust the magnitude of the weights.
[W] Check verbose logs for the list of affected weights.
[W] - 224 weights are affected by this issue: Detected subnormal FP16 values.
[I] Finished engine building in 1063.897 seconds
[I] Saving engine to models/accelerate/tensorrt/gsdf/Counterfeit-V2.5/engine/unet.plan

[l-dawei]

Hello, could you use dynamic shape?

[Isotr0py]

@l-dawei Yes, dynamic shape can work normally.