Android rpc tries to connect to the machine running tracker (btw in its turn
tracker will try to connect to the phone as well during the work). Android RPC
has special fields which you have to fill before you switch "Enable RPC"
slider. Phone must be able to connect tracker machine. The simple
Do you point real ip address of machine running tracker in the android rpc?
Android and machine running tracker should be in the same subnet because phone
has to connect to that machine to the port 3939 in your case.
---
[Visit
Topic](https://discuss.tvm.apache.org/t/can-not-connect-andri
[quote="namduc, post:1, topic:10908"]
`relay.vm.compile`
[/quote]
virtual machine execution cannot be tuned so far
---
[Visit
Topic](https://discuss.tvm.apache.org/t/how-to-apply-best-history-after-auto-scheduler-for-relay-vm-compile/10908/2)
to respond.
You are receiving this because you
[quote="namduc, post:8, topic:10889"]
what are the “fist lines” you mentioned here !?
[/quote]
I see in quote a part of the tuning trace starting from `ElapsedTime(s) 17787
EstimatedLatency(ms)471.479 Trials 10176`, it refers to 10899 trial, and I
referred it as "first line". while if yo
[quote="namduc, post:6, topic:10889"]
after that i got the same result with a model onnx running on onnxruntime
[/quote]
It might be that onnxruntime was able to use hardware resources the most
efficient way and improving of the inference time more is possible but might be
hard. And TVM get the
@myproject24 did you have a chance to verify if passing of CPU NDArray to input
and using of the other API for copy data helps to solve the problem or not?
---
[Visit
Topic](https://discuss.tvm.apache.org/t/irrespective-of-input-same-output-ios-tvm-model/10765/11)
to respond.
You are rec
in
[MetalWorkspace::CopyDataFromTo](https://github.com/apache/tvm/blob/main/src/runtime/metal/metal_device_api.mm#L228)
three situation is handled:
1. Copy from Metal to Metal
2. Copy from CPU to Metal
3. Copy from Metal to CPU
I.e. in your case one more extra copy will be done and it will be m
[quote="myproject24, post:7, topic:10765"]
`w*h*c`
[/quote]
missing of `*4` - is it a problem of copy-past you forgot to compensate the
size of the float? since TVMArrayCopyFromBytes deal with bytes, not floats.
Another question why do you need m_gpuInput? you can use only NDArray been
alloca
[quote="myproject24, post:5, topic:10765"]
m_gpuInput is in metal context **kDLMetal**
[/quote]
How do you populate data to this NDArray?
---
[Visit
Topic](https://discuss.tvm.apache.org/t/irrespective-of-input-same-output-ios-tvm-model/10765/6)
to respond.
You are receiving this because
which context do you use for creation of `m_gpuInput` and how do you populate
data?
BTW, you are loading params and json that means that you are using packed zip
format for storing model. Is it more convinient for you? Does it find artefacts
during the loading of model? I prefer to use everyt
Autoscheduler completely ignores manually developed schedules, it creates new
schedules using only compute function and its Ansor rulls how to transforms one
schedule to another. You will not be able to get default scheduling through
Ansor workflow. Instead of this it might be better to take i
You need to run install_name_tool for your model library like
```
install_name_tool -id @rpath/model.dylib model.dylib
```
and then just point the name of the library without additional path in iOS app
like
```
m_mod = tvm::runtime::Module::LoadFromFile("model.dylib");
```
---
[Visit
Top
One small comment in addition to previous Egor's message - there should be one
more option for cmake command line added -
`-DCMAKE_CXX_FLAGS="-fembed-bitcode-marker"`
---
[Visit
Topic](https://discuss.tvm.apache.org/t/compile-pytorch-model-for-ios/9995/6)
to respond.
You are receiving t
13 matches
Mail list logo
