Training and rollout logprobs matched bit for bit on ROCm. The @RLKernel team integrated RL-Align/RL-Kernel with vllm-project/vime. A 200-step Qwen3-8B GRPO run on 8× @AMD MI300X recorded zero logprob mismatches between Megatron training and vLLM rollout. The strict path aligns reduction order, intermediate precision, rounding points, and math primitives across both sides. Deep dive:

Not the ship rollout we were expecting! 🚀 While Ship 41 prepares to rollout to the launch site we were surprised with a nosecone heading to MegaBay 2 for stacking, the production speed is insane! 📸: @RobertsCraneArt For WAI Media

Starlink’s rollout in Vietnam is expanding again 🇻🇳 Da Nang has approved Starlink deployment across 14 remote locations: • 4 villages • 10 school sites • Installation from Sept. 21–30 • 6 months of Starlink service provided free The rollout is focused on mountainous areas where conventional telecom coverage is weak or difficult to reach The four villages are Aur, Apát, Đắc Chơ Đây and Thôn 3 Another real-world example of Starlink reaching places terrestrial networks still struggle to cover

BREAKING: Qatar Airways’ Starlink rollout has reached insane scale. The airline now operates up to 340 Starlink-enabled flights EVERY DAY across 150+ widebody aircraft. Since launching Starlink in October 2024: • 23M+ passengers connected • 86,000+ Starlink flights • Speeds up to 500 Mbps • Boeing 777, Airbus A350 and Boeing 787 equipped • Long-haul connectivity across six continents Qatar Airways now has the world’s largest Starlink-equipped widebody fleet. 340 flights every single day. Starlink is becoming the new standard for internet in the sky.

Qatar Airways’ Starlink rollout has reached insane scale The airline now operates up to 340 Starlink-enabled flights EVERY DAY across 150+ widebody aircraft Since launching Starlink in October 2024: • 23 MILLION+ passengers connected • 86,000+ Starlink flights • Speeds up to 500 Mbps • Boeing 777, Airbus A350 and Boeing 787 aircraft equipped • Long-haul connectivity across six continents Qatar Airways now operates the world’s largest Starlink-equipped widebody fleet 340 flights every single day.... Starlink is becoming the new standard for internet in the sky

“FlashREINFORCE: Critic-Free Single-Rollout Asynchronous RL for Agentic LMs” This paper argues that scalable agent RL doesn’t need multiple rollouts of the same prompt, but can learn from each trajectory as soon as it finishes. They combine reward centering across independent prompts, sequence-level trust regions for stale trajectories, and sample-mean optimization so long failures don’t dominate training, creating stable critic-free learning from just one rollout per prompt. This moves from group-based RL that spends compute repeatedly sampling the same prompts to asynchronous RL where every rollout covers a new prompt, immediately becomes training data, and lets agents learn from more diverse experiences with less rollout compute.