A periodically asynchronous on-policy RL system for LLM post-training achieves up to 3x throughput gains by separating inference and training with periodic policy synchronization and a tri-model architecture.
Mindspeed rl: Distributed dataflow for scalable and efficient rl train- ing on ascend npu cluster,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Periodic Asynchrony: An On-Policy Approach for Accelerating LLM Reinforcement Learning
A periodically asynchronous on-policy RL system for LLM post-training achieves up to 3x throughput gains by separating inference and training with periodic policy synchronization and a tri-model architecture.