reminisce commented on issue #14040: Reformat of TensorRT to use subgraph API
URL: https://github.com/apache/incubator-mxnet/pull/14040#issuecomment-461542427
> Also an argument for onnx-tensorrt is that there is more Ops supported
with plugins implemented (slice, some activation, resize
reminisce commented on issue #14040: Reformat of TensorRT to use subgraph API
URL: https://github.com/apache/incubator-mxnet/pull/14040#issuecomment-461539504
> weights for TensorRT node are on CPU while the rest of the graph is on GPU.
This is not true. When binding completes, you
reminisce commented on issue #14040: Reformat of TensorRT to use subgraph API
URL: https://github.com/apache/incubator-mxnet/pull/14040#issuecomment-461291971
> @reminisce
> you can't mix context in a single graph
What do you mean "mix context"? We only have one context which gpu in
reminisce commented on issue #14040: Reformat of TensorRT to use subgraph API
URL: https://github.com/apache/incubator-mxnet/pull/14040#issuecomment-461280671
@Caenorst
Just want to clarify, I'm not blocking this PR. We can think through about
these comments and make incremental
reminisce commented on issue #14040: Reformat of TensorRT to use subgraph API
URL: https://github.com/apache/incubator-mxnet/pull/14040#issuecomment-460792134
Great to see this is happening. I have two high-level comments:
1. If you use the subgraph API, there should be no needs to add