[GH-ISSUE #159] bug (memory leak & OOM) New comfy aimdo (comfy compiler pr) conflicts with New Blockspare attention node PR (from kijay) - MINIMAX H3 VIDEO GENERATION #49

New issue

Closed

opened 2026-10-26 22:19:42 +00:00 by bsakaguchi · 1 comment

bsakaguchi commented

2026-10-26 22:19:42 +00:00

Owner

Originally created by @bhusann on GitHub (Sep 6, 2026).
Original GitHub issue: https://github.com/Comfy-Org/comfy-kitchen/issues/159

https://github.com/Comfy-Org/ComfyUI/issues/16144

Originally created by @bhusann on GitHub (Sep 6, 2026). Original GitHub issue: https://github.com/Comfy-Org/comfy-kitchen/issues/159 https://github.com/Comfy-Org/ComfyUI/issues/16144

bsakaguchi closed this issue

2026-10-26 22:19:42 +00:00

bsakaguchi commented

2026-10-26 22:19:43 +00:00

Author

Owner

@bhusann commented on GitHub (Sep 6, 2026):

https://github.com/Comfy-Org/ComfyUI/pull/16148

This PR fixed OOM

And the memory usage of this new aimdo is great too. I test with NO vram headroom and NO reserve vram and NO lowvram arguments, and it does not crash and used my vram to the fullest utilisation and also I observed less disk read activity during steps too

(My specs: 8gb vram + 16gb ram, model is minimax fl2va w4a8 model)

@bhusann commented on GitHub (Sep 6, 2026): https://github.com/Comfy-Org/ComfyUI/pull/16148 This PR fixed OOM And the memory usage of this new aimdo is great too. I test with NO vram headroom and NO reserve vram and NO lowvram arguments, and it does not crash and used my vram to the fullest utilisation and also I observed less disk read activity during steps too (My specs: 8gb vram + 16gb ram, model is minimax fl2va w4a8 model)

bsakaguchi referenced this issue

2026-10-26 22:20:25 +00:00

[PR #49] [MERGED] Add optimized adaln implementation. #103