Closed lingwei-gu closed 2 years ago
@lingwei-gu hi lingwei! Thanks for helping updating the repro doc. tho the monobert r expected to hv MRR@10 to be 0.35+. chould u chime in the issue #200 and see if the decay
line is also missing from the script u were using as well?
@crystina-z You are right it is missing reranker.trainer.decay=0.1
in this file: config_msmarco.txt
I'll run the experiment again and see how it goes
@crystina-z The decay
line was no longer missing after using the latest nimasadri11/capreolus
repository.
Submitted the job through SLURM on Compute Canada.
One major issue for me in the sample_slurm_script.sh was: After loading module StdEnv/2018.3, module java/11 became inactive and cannot be found anymore. Thus, I looked up the documentations and changed it to StdEnv/2020, then it works.