title: the same run with the engine memory share lowered from 0.4 to 0.2
source: manuals/miles-101/capture/gpu/out-gb10/ray-oom-4b-lowmem.log
lang: text
note: Compare the node memory and the trainer process with the first kill.
---
Run e4-4b-lowmem on 2026-10-22: Qwen3-4B, --sglang-mem-fraction-static 0.2, default Ray memory threshold.
(TrainerController pid=2518) ray.exceptions.OutOfMemoryError: 1 worker(s) were killed due to the node running low on memory. Memory on the node (IP: 172.17.0.3, ID: 52a2f7678ec4002fb0c757192d14dfb4606f3204b11008afe40fe377) was 119.18GB / 121.69GB (0.979391)
(TrainerController pid=2518) Top 10 memory users: PID	MEM(GB)	COMMAND
(TrainerController pid=2518) 2523	36.75	ray::FSDPTrainRayActor.train
(TrainerController pid=2518) 3259	2.67	sglang::scheduler
(TrainerController pid=2518) 2515	1.32	ray::RolloutExecutor
