Expired60% confidenceFactTime unknown
Cursor and Fireworks adopted an asynchronous pipeline solution for RL training, keeping all GPUs running at full capacity at all times.
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
5/31/2026
First Seen
Valid until: 8/29/2026(expired)
Sources
Related Entities
Related Claims
UnverifiedCursor and Fireworks jointly developed Router Replay technology, which resolves the problem of inconsistent expert routing between the inference side and the training side of MoE models during RL training.73% similarUnverifiedRL训练涉及环境交互（CPU密集）和策略优化（GPU密集）两个阶段的交替66% similarUnverifiedCursor borrows production inference GPUs during low user traffic periods to accelerate training.64% similarUnverifiedGrok 4.5's training incorporated tens of trillions of Cursor usage data and was combined with an RL training pipeline.63% similarUnverified梯度同步是指数据并行训练中每个GPU通过Ring-AllReduce算法将梯度与所有其他GPU聚合的操作，其通信量与模型参数量成正比63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/5804API
curl https://kongchang.com/api/v1/knowledge/claims/5804MCP
get_claim(id=5804)