mirror of
https://github.com/comfyanonymous/ComfyUI.git
synced 2026-03-10 03:37:33 +08:00
9250191c65
DynamicVRAM's on-demand model loading/offloading conflicted with process isolation in three ways: RPC tensor transport stalls from mid-call GPU offload, race conditions between model lifecycle and active RPC operations, and false positive memory leak detection from changed finalizer patterns. - Marshal CUDA tensors to CPU before RPC transport for dynamic models - Add operation state tracking + quiescence waits at workflow boundaries - Distinguish proxy reference release from actual leaks in cleanup_models_gc - Fix init order: DynamicVRAM must initialize before isolation proxies - Add RPC timeouts to prevent indefinite hangs on model unavailability - Prevent proxy-of-proxy chains from DynamicVRAM model reload cycles - Add torch.device/torch.dtype serializers for new DynamicVRAM RPC paths - Guard isolation overhead so non-isolated workflows are unaffected - Migrate env var to PYISOLATE_CHILD |
||
|---|---|---|
| .. | ||
| apis | ||
| util | ||
| __init__.py | ||
| nodes_bfl.py | ||
| nodes_bria.py | ||
| nodes_bytedance.py | ||
| nodes_elevenlabs.py | ||
| nodes_gemini.py | ||
| nodes_grok.py | ||
| nodes_hitpaw.py | ||
| nodes_hunyuan3d.py | ||
| nodes_ideogram.py | ||
| nodes_kling.py | ||
| nodes_ltxv.py | ||
| nodes_luma.py | ||
| nodes_magnific.py | ||
| nodes_meshy.py | ||
| nodes_minimax.py | ||
| nodes_moonvalley.py | ||
| nodes_openai.py | ||
| nodes_pixverse.py | ||
| nodes_recraft.py | ||
| nodes_rodin.py | ||
| nodes_runway.py | ||
| nodes_sora.py | ||
| nodes_stability.py | ||
| nodes_topaz.py | ||
| nodes_tripo.py | ||
| nodes_veo2.py | ||
| nodes_vidu.py | ||
| nodes_wan.py | ||
| nodes_wavespeed.py | ||