Tal
Tal
open-source models are dramatically inferior in code, compared to GPT4. they may work, and may not, since they fail to understand and obey the prompt instructions. If GPT4 does not...
llama3 (which is a bad model for code) is failing to obey the prompt. it's not an 'error', but more like showing that this model is not good enough. share...
@bdelgado1995 thanks, i will review it. Maybe with some post-processing of the answer it can still work
@bdelgado1995 i updated the post-processing script, and I was able to successfully use llama-8b and llama-70b on the baseline script (be sure to use the updated the repo)
p.s. what do i mean by "not obeying instructions":  
@benedict-lee i like this PR. the code is cleaner with it 👍
/improve
@davidparry why did you close this PR ?
> Correct the calculation of coverage percentage by multiplying by 100 > > The coverage_percentage calculation in parse_coverage_report_jacoco should multiply covered by 100 to get the percentage value. > cover_agent/CoverageProcessor.py...
@gitworkflows can you share a place\reference where this practice is used ?