Part 2 of the 120B model resilience test. After the phone dropped offline with zero downtime, I tried harder to break it — closed the lid on one of the Macs holding 7.4GB of the model and expected the whole cluster to destabilize.
It didn't. In fact, I found something I didn't expect: even though that Mac stopped sending its heartbeat to the coordinator, its inference process was still running and still reachable — a "ghost node" quietly still doing its job while showing as offline on the dashboard. RAMDeck had already worked out how to redistribute that shard to the old Intel MacBook if I approved the rebalance, but the model never actually needed it to.
Down to 4 actively-heartbeating devices, still healthy, still answering prompts.
RAMDeck — Pool Your RAM. Run Bigger AI.
Run and train models too big for any single machine, using the PCs, Macs, mini PCs and phones you already own.
Leave a Reply