The goal is to port LiT5-Distill and LiT5-Score over to RankLLM.
Completed
Synced the LiT5 integration structure with the latest master, from command-line execution to the rankllm class
Built RankFiDDistill(RankLLM) and RankFiDScore(RankLLM) in rank_fid.py
Built LiT5DistillReranker and LiT5ScoreReranker in lit5_reranker.py
Next Steps
Inherited functions in RankFiDDistill, like create_prompt and run_llm, still need adjustments and testing to ensure LiT5 performs correctly
Implemented inherited functions in RankFiDScore
Validate the process of using LiT5 models within RankLLM, ideally we have a central command so run LiT5 as python src/rank_llm/scripts/run_rank_llm.py with --args
Reference URA-Project
Issue 35 RankLLM: Merge LiT5 Models into RankLLM https://github.com/castorini/ura-projects/issues/35
The goal is to port LiT5-Distill and LiT5-Score over to RankLLM.
Completed
Next Steps
python src/rank_llm/scripts/run_rank_llm.pywith --args