Nodes/AllmaNodes/Clear Allma VRAM
ComfyUI Node

Clear Allma VRAM

Unloads the Allma backend and frees its GPU, so the card can be used by ComfyUI. Pass-through: wire it anywhere and it fires at that point in the run.

By OliveiraNickolas·Created 4 days ago·Updated a day ago· 0
Clear Allma VRAM
  • any
  • connectivity
  • any
  • status
kill_orphanstrue
wait_until_freetrue
timeout90
enabledtrue
CategoryAllma/utils

Inputs (6)

NameTypeDefaultDescription
any*Anything at all — returned unchanged. This is the trigger: the cleanup happens when this value is needed.
connectivityoptALLMA_CONNECTIVITYWhere the Allma server lives. Without it, 127.0.0.1:9000 is assumed.
kill_orphansoptBOOLEANtrueAlso terminate inference-backend processes still holding VRAM after the unload — the ones Allma has lost track of. Only processes whose command line names a known backend (llama-server, vllm, sglang) are ever signalled; ComfyUI is never touched. OFF: only the polite unload is attempted.
wait_until_freeoptBOOLEANtrueHold the graph until the memory is actually released, so the next node does not start loading into VRAM that is still occupied.
timeoutoptINT905–600Seconds to wait before giving up and letting the graph continue anyway.
enabledoptBOOLEANtrueOFF: pure pass-through, nothing is freed. Lets you keep the node wired while iterating.

Outputs (2)

NameTypeDescription
any*
statusSTRING