Hi all,
Been test driving deepseek v4 flash the last couple of days, but decided to revert back to qwen 3.5 122B and qwen 3.6 35b combo
Running my previous recipe with with the mod gpu-mem-util-gb, the script fails to execute, “2 out of 3 hunks FAILDED”.
Removing the mod, now vllm fails with “unknown unrecognized arguments: --gpu-memory-utilization-gb”
Anyone know which last version was working and how to get it back?
@eugr, thoughts?