stage11 stage13 issues
Wujianting opened this issue · 9 comments
Hi @Wujianting,
For the first problem:
You can try to add to env.sh
:
export PYTHONPATH=<your path>/anonymization_metrics:$PYTHONPATH
This is done in
Ideally, this should be in
env.sh
For the second problem, could you please attach the corresponding logs rescorelm.*.log
?
对于第二个问题,能否附上相应的日志
rescorelm.*.log
?
rescorelm.1.log
rescorelm.2.log
rescorelm.3.log
rescorelm.4.log
rescorelm.5.log
rescorelm.6.log
rescorelm.7.log
rescorelm.8.log
rescorelm.9.log
rescorelm.10.log
rescorelm.11.log
rescorelm.12.log
rescorelm.13.log
rescorelm.14.log
rescorelm.15.log
rescorelm.16.log
That's all
Thank you for your help.
Hi @Wujianting, For the first problem:
You can try to add to
env.sh
:export PYTHONPATH=<your path>/anonymization_metrics:$PYTHONPATH
This is done in
Ideally, this should be in
env.sh
I have successfully solved the problem of stage11. Thank you for your help.
Solution: I used git clone to download anonymization again_ metrics.
But I still haven't solved stage13. Please help me solve it in your spare time. Thank you very much.
For the second problem, some of the processes were killed by the system as written in the logs, i.e., in rescorelm.1.log
: 31838 Killed
对于第二个,一些被系统处理的问题,如日志中所写,
rescorelm.1.log
:31838 Killed
Hello, let me make it clear that the second screenshot is the result of running stage13 only. Is there a necessary process for stage13 in the previous stage?
Do you mean that the whole process of stage 1-16 can not run continuously?
To run Stage 13
, Stage 12
(Make ASR evaluation subsets) should be completed; Stage 11
is not necessary for that.
In Stage 13
, most of your 16 processes correctly worked, and some of them were killed by another process/service in your system. I can not say which service in your system and why killed some of the processes at Stage 13
. Where did you run your scripts - on the cluster or work station?
What happens with the processes in wsl2 if the memory exceeds the available capacity? Would it kill the processes?
You can try to run decoding (and LM scoring) in 1 thread instead of 16.