AlphaUncommon
qiqihezh/agentic-grpo-longhorizon
Fixing GRPO training collapse in long-horizon multi-tool agents. A lightweight PRM-Lite + LATA joint approach achieves +37% over vanilla GRPO on τ-bench airline (50-task, multi-turn).
- Owner
- qiqihezh (personal account)
- Stars
- Stars: 266
- Forks
- Forks: 17
- Language
- Python
- License
- None
- Created
- 2026
- Topics
- agentic-aigrpolong-horizonprocess-reward-modelqwenreinforcement-learningtau-benchtool-callingmulti-turn-agents
- Tier
- Alpha (250 to 699 ★) What this tier means
More repos by qiqihezh
All 2 repos by qiqihezh- qiqihezh/deepresearch-agentStars: 175PythonPrototype
Nearby in Python
All Python repos- julianmb/q38rocmStars: 266PythonAlpha
- Gen-Verse/RecurisStars: 266PythonAlpha
- zjwfufu/SMDStars: 266PythonAlpha
- kassoulet/soundconverterStars: 266PythonAlpha
- erezsh/plyplusStars: 266PythonAlpha
- ClusterLabs/pcsStars: 266PythonAlpha