Dheeraj Peri

@yachty66 Did you try with the latest main branch or the release 2.6 ?

1) Can we perform slice on dynamic dimensions ? 2) I'm facing slice layer conversion issues on llama2 Here's the reproducer : 1) Please login via huggingface-cli (install it via...

@apbose I'm seeing the following error ```py File "/home/dperi/Downloads/TensorRT/py/torch_tensorrt/dynamo/conversion/converter_utils.py", line 530, in convert_with_type_enforcement return func(ctx, target, new_args, new_kwargs, name) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ File "/home/dperi/Downloads/TensorRT/py/torch_tensorrt/dynamo/conversion/aten_ops_converters.py", line 884, in aten_ops_slice return impl.slice.slice_op( ^^^^^^^^^^^^^^^^^^^^ File...

@zewenli98 1) Having it in op_evaluators is fine due to its duality 2) How are you passing static shape ITensor to the converter ? Can you show your testcase ?...

`torch._dynamo.mark_dynamic(input_ids, 1, min=7, max=1023)` fixes this issue though. Previously, I was using `min=2`

Pytorch issue : https://github.com/pytorch/pytorch/issues/125604

```py torch.fx.experimental.symbolic_shapes.ConstraintViolationError: Constraints violated (L['input_ids'].size()[1])! For more information, run with TORCH_LOGS="+dynamic". - Not all values of L['input_ids'].size()[1] = L['input_ids'].size()[1] in the specified range L['input_ids'].size()[1]

One idea from Angela to avoid overhead of setting up engines on windows is to handle in `torch_tensorrt.load()`. This modifies and creates a new graph which makes the engines go...

Proposal: 1) `torch_tensorrt.dynamo.cross_compile` - Input is exported program, output is an exported program. 2) Exporter: a) Currently, for x86 compilation, we create a custom op `execute_engine(inputs, TRTEngine)` in the graph...

Thanks for reporting this issue @LinzhouLi It does seem like a fix in Pytorch would be the better solution here. Let me check on that. Here is a workaround for...