Warning DPO merges are contaminated with test data from benchmarks this should be flagegd!
#4
by
mishaml77
- opened
For the people out there , don't trust efficiency of this model , as its based on dubious models that used contaminated training dataset to boost their performance on the openllm leaderboard.
For the people out there , don't trust efficiency of this model , as its based on dubious models that used contaminated training dataset to boost their performance on the openllm leaderboard.
How can you tell which are contaminated @mishaml77 ?