Williams et al.
Splitting an RLVR budget across multiple adapters outperforms a single adapter for majority voting
Training multiple LoRA adapters on disjoint shards of the data preserves the diversity needed for effective voting, while a single concentrated adapter loses vote accuracy despite improving pass@1.
Top University
Jonathan Williams · Esin Tureci Karthik R. Narasimhan
Princeton University
Research Digest··3 min read
The authors show that standard RLVR training, while improving single-sample accuracy, increases error correlation among samples, degrading majority vote.
Why this paper
From Princeton University
In one line
Splitting an RLVR training budget across multiple LoRA adapters improves majority vote accuracy over training a single adapter.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (2 noted)
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§