The authors built LoongSage on top of Megatron for distributed training and SGLang for high-throughput inference.
Agentic RL framework keeps asynchronous training stable with bounded staleness
LoongSage's lightweight, agent-native design couples Megatron and SGLang to support asynchronous rollouts, topology-aware weight transfer, and multi-teacher distillation in four validated training runs.
Independent
LoonSage Team · Bowei Song · Caiwei Zhao · Chuwen Liu · Dechuang Gu · Dong Fang · +42 more
Research Digest··2 min read
The LoongSage team presents a production-oriented reinforcement learning framework for agentic LLM training, with a roughly 14,000 line core built around fully asynchronous sampling and optimization.
Why this paper
Independent
In one line
LoongSage makes agentic LLM reinforcement learning more efficient through asynchronous execution, bounded trajectory staleness, topology-aware weight transfer, and train-inference consistency.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§