Hi,
I can run Qwen3-MoE models (nKvHeads: 4) on 4 nodes without problems and it works great - thank you very much!
If I try to run the same model on 8 nodes I get this error:
🚨 Critical error: This version does not support more nodes than the number of KV heads in the model
It seems there can't be more than nKvHeads nodes. Is there any way to run Qwen3-MoE models on 8 nodes?