fix(qwen): make room in VRAM for the quantized Qwen2.5-VL encoder and release it afterwards (#9147) - #9572

Merged

lstein merged 9 commits into

invoke-ai:mainfrom

lstein:fix/qwen-quantized-encoder-vram

Sep 15, 2026

Commits on Sep 7, 2026

fix(qwen): make room in VRAM before loading the quantized Qwen2.5-VL encoder

lstein
and
claude
committed
7d07921
fix(qwen): release the quantized encoder before emptying the CUDA cache

lstein
and
claude
committed
aeb0b53
fix(qwen): make_room_in_vram is a no-op on a CPU execution device; tighten tests

lstein
and
claude
committed
a781393
fix(qwen): release the encoder on the error path and drop activations before empty_cache

lstein
and
claude
committed
42eed4b
docs: regenerate invocation-context data for make_room_in_vram

lstein
and
claude
committed
84c598d