cjluo
1
When I try to re-serialize my model after upgrading tensorrt from 3 to 4, I got the following error:
…/builder/costTensor.h:271: nvinfer1::builder::ChoiceTupleMap::ChoiceTupleMap(unsigned int, T) [with T = float]: Assertion `rank <= ChoiceTuple::kMAX_RANK’ failed.
It probably fails in a customized plugin. Could anyone help interpret what this error means?
I’m using cuda 8.0
Thanks!
cjluo
2
Call stack:
*** SIGABRT (@0x3e900001dbc) received by PID 7612 (TID 0x7f94355f9b00) from PID 7612; stack trace: ***
@ 0x7f940c4ae390 (unknown)
@ 0x7f94034e9428 gsignal
@ 0x7f94034eb02a abort
@ 0x7f94034e1bd7 (unknown)
@ 0x7f94034e1c82 __assert_fail
@ 0x7f94137b99c0 (unknown)
@ 0x7f94137bc879 nvinfer1::builder::chooseFormatsAndTactics()
@ 0x7f94137c78a9 nvinfer1::builder::makeEngineFromGraph()
@ 0x7f94137cb79d nvinfer1::builder::buildEngine()
…
Arthes
3
cuda-9.0
tx2
tensorrt so 4.1.3, include 4.0.2(jetpack install)
…/builder/costTensor.h:271: nvinfer1::builder::ChoiceTupleMap::ChoiceTupleMap(unsigned int, T) [with T = float]: Assertion `rank <= ChoiceTuple::kMAX_RANK’ failed.
model works ok on P100,cuda-9.0
Arthes
4
On TX2, I use tensorrt so 4.0.4 and there is no error…Confused.
Hello, I met the same problem with you when I upgrade tensorrt 3 to 4. Have you solved it?