GRPO produces modest math gains with small knowledge loss, while SFT gives larger math gains but degrades knowledge benchmarks more, with parameter analyses suggesting the loss comes from larger mid-layer MLP updates.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Scalpel vs. Hammer: GRPO Amplifies Existing Capabilities, SFT Replaces Them
GRPO produces modest math gains with small knowledge loss, while SFT gives larger math gains but degrades knowledge benchmarks more, with parameter analyses suggesting the loss comes from larger mid-layer MLP updates.