Prime Intellect has released prime-rl 0.6.0, an open framework for asynchronous reinforcement learning on trillion-parameter Mixture-of-Experts models. It trained GLM-5 on SWE tasks at up to 131k sequence length, with sub-5-minute step times and 256 rollouts, on 28 H200 nodes. This breakdown covers the inference and training optimizations behind those numbers — FP8 inference, Wide Expert Parallelism, prefill/decode disaggregation, router replay, and 3-D parallelism (FSDP, EP, CP).
The post Prime Intellect Releases prime-rl 0.6.0 to Train Trillion-Parameter MoE Models on Agentic RL Workloads appeared first on MarkTechPost.
source & further reading
marktechpost.com — original article
Gradium Launches stt-translate and s2s-translate, Real-Time Speech Translation Models Beating gpt-realtime-translate on Accuracy and Latency
How to Design an OpenHarness Style Agent Runtime with Tools, Memory, Permissions, Skills, and Multi-Agent Coordination
Using Graphify and NetworkX to Map Python Codebase Structure with God Nodes, Communities, and Architecture Visualizations