Why do we have to run merge?

[ramsey-coding]

Can't I just generate the the edit locations and then run the repair?

mkdir results # where we will save our results
python agentless/fl/localize.py --file_level --related_level --fine_grain_line_level \
                                --output_folder results/location --top_n 3 \
                                --compress \
                                --context_window=10 \
                                --temperature 0.8 \
                                --num_samples 4

Why I have to also run the following command?

python agentless/fl/localize.py --merge \
                                --output_folder results/location_merged \
                                --start_file results/location/loc_outputs.jsonl \
                                --num_samples 4

What is it doing really? @dengyinlin @brutalsavage

[brutalsavage]

Hi @ramsey-coding these commands are to reproduce our exact setup we used in the paper on SWE-Bench Lite.

In this example the first command will generate 4 sets of edit locations (using sampling) and the second command will proceed to merge these edit locations together to do pair wise merging (please see the our paper and readme for a bit more detail)

If you prefer to generate just the one set of edit locations (without having to merge), you can look at our top-level readme or simply remove the --temperature and --num_samples from the first command and do not run the second command