Reference library
Training Verifiers to Solve Math Word Problems
Paper summary
Introduces GSM8K mathematical word problems and trains verifiers to rank generated solutions. Sampling several candidates and selecting by a learned verifier improves evaluated performance over direct fine-tuning baselines.
Relevance to the project
A comparison for separate behavior generation and evaluation mechanisms. Independent checks are needed because a learned verifier can be systematically mistaken.
Verifier-selected success depends on labels, candidate budget, and benchmark distribution.
Source record
- Reference ID
- gsm8k
- Archived version
- 2110.14168v2
- Summary provenance
- reference-library/papers/gsm8k/metadata.json · summary
- Review status
- initial orientation summary; not a critical full-paper review
- Reading scope
- Selected abstract/opening and available closing discussion/conclusion passages in the preserved text extract; broader project notes provide context. No reproduction of experiments, exhaustive methods review, or equation/figure audit.