Open larochef opened 1 year ago
Also Joining the system core dump log if it can help core_dump.log
I can reproduce on ubuntu 22.04.
There isn't much we can do on DJL side. This is related to torch.jit LLVM compiler. There are a few discussions on pytorch forums, but there isn't a clear solution.
I could git rid of the error using the -Dai.djl.pytorch.graph_optimizer=false
parameter. I'm not sure of all the implications, but it will always be better than a crashing app
I could git rid of the error using the
-Dai.djl.pytorch.graph_optimizer=false
parameter. I'm not sure of all the implications, but it will always be better than a crashing app
Thanks for providing this workaround. graph_optimizer
can improve throughput for certain models. See: https://docs.djl.ai/master/docs/development/inference_performance_optimization.html#graph-executor-optimization
We actually observed several negative impact with graph_optimizer turned on. It looks like it triggered torch.jit
bug in your case.
Description
When running djl with javafx on linux x64 (archlinux) using the cpu flavor, I get the following error when I try to get embeddings twice from a model. If I run the same code with cuda, it runs fine. It seems like it does'nt happen on windows, but can happen with WSL.
Expected Behavior
system is expected to continue without error
Error Message
How to Reproduce?
(If you developed your own code, please provide a short script that reproduces the error. For existing examples, please provide link.)
I created a minimal repo to reproduce it: https://github.com/larochef/javafx-djl
Steps to reproduce
(Paste the commands you ran that produced the error.)
./gradlew run
What have you tried to solve it?
Environment Info
Please run the command
./gradlew debugEnv
from the root directory of DJL (if necessary, clone DJL first). It will output information about your system, environment, and installation that can help us debug your issue. Paste the output of the command below: