TVM simple tune hang

Troubleshooting
#1

Hello there,

I am new to TVM and try to run a simple tuning according to Compiling and Optimizing a Model with TVMC — tvm 0.11.dev0 documentation, but the tune hang there as below, (compile and run works fine w/ llvm target, just tune get hang…)

Any comments?

python -m tvm.driver.tvmc tune --target “llvm -mcpu=broadwell” --output resnet50-v2-7-autotuner_records.json --tuner random --number 1 --repeat 1 resnet50-v2-7.onnx

tvm\target\target.py:281: UserWarning: target_host parameter is going to be deprecated. Please pass in tvm.target.Target(target, host=target_host) instead. "target_host parameter is going to be deprecated. "

[Task 1/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/40) | 0.00 s

Hang. no further print… And CPU utilization dropped which seems no more workload is running…

Any comments/suggestions?

Thanks in advance!

1 Like
#2

It is running on Windows, does anyone encounter the same issue?

1 Like
#3

Can you try adding -vvv to add verbosity to the output, and perhaps we can get a clue of what’s happening?

#4

Hi leandron,

“-vvv” seems not work, I use $env:TVM_LOG_DEBUG=“DEFAULT=0”, and get a lot of warning log as below,

tvm\src\arith\int_set.cc:521: Warning: cannot evaluate set type tir.Call

Thanks, Larry

#5

set “export TVM_LOG_DEBUG=0” in your environment may make it disappear. it works for me.

#6

I am new to TVM and I am experiencing the same issue. I was able to run with the -vvv option as suggested in the thread and this is my output:

$ python3 -m tvm.driver.tvmc tune --target “llvm” --output resnet50-v2-7-autotuner_records.json resnet50-v2-7.onnx -vvv DEBUG:TVMC:target input is plain text: llvm INFO:TVMC:Default --min-repeat-ms for this target is 0 INFO:TVMC:Starting localhost tuning. INFO:TVMC:Selected 25 tasks for tuning. INFO:TVMC:Autotuning with 40 trials per task. INFO:TVMC:Autotuning with configuration: {‘tuner’: ‘xgb’, ‘trials’: 40, ‘early_stopping’: None, ‘measure_option’: {‘builder’: <tvm.autotvm.measure.measure_methods.LocalBuilder object at 0x7f1dd3ed2650>, ‘runner’: <tvm.autotvm.measure.measure_methods.LocalRunner object at 0x7f1d81fa9990>}, ‘tuning_records’: None} [Task 1/25] Current/Best: 0.00/ 0.00 GFLOPS | Progress: (0/40) | 0.00 s

In my case, I compiled TVM from scratch and installed LLVM binaries. I am running in Ubuntu 22.04.3 LTS running inside Windows using WSL. Don’t know if that could be the issue since from what I read autotune uses some kind of server to do the search. I am able to compile without issues. Any ideas or help will be appreciated.