Tool use / function calling

GRPO

Reinforcement Learning Outperforms Supervised Fine-Tuning: A Case Study on Audio Question Answering

Superseded baseline#6 of 55 most-superseded · first seen Mar 14, 2025

Superseded — cited as a baseline and beaten by newer methods

2 papers critique it · 1 beat it on benchmarks

What papers say

Verbatim critique sentences, each from a paper that cites GRPO as a baseline.

GRPO fine-tunes via weight-space RL but requires thousands of rollouts.
Reinforced Agent: Inference-Time Feedback for Tool-Calling Agents
However, their learning signal fundamentally depends on within-group variability: if the rewards within a sampled group have near-zero or even completely zero standard deviation, the group-normalized advantage becomes degenerate and policy updates vanish.
RC-GRPO: Reward-Conditioned Group Relative Policy Optimization for Multi-Turn Tool Calling Agents

Beaten on benchmarks

Head-to-head results where a newer method reports beating GRPO. Values are copied from the source paper's tables — verify against the cited paper.

What to use instead

Recent methods in the same sub-problem, not yet superseded in the knowledge base — arXiv benchmark leaders, not vetted production recommendations.