From Reweighting to Rewriting: Unlocking the Intervention Effects of Influential Samples in Training Data Attribution
Open paper
Benchmark
Model
Rank
Results
arithmetic-reasoning-on-gsm8k
OLMo2-1B + influence rewriting (Opposed Helpful)
–
Accuracy: 37.07
sentence-completion-on-hellaswag
OLMo2-1B + influence rewriting (Opposed Harmful)
–
Accuracy: 62.93
Rank counts only results with a code link.