Claude a13f037f72 H3 training: modulate blocks out of place so the backward pass stops keeping every block's recompute alive ...

Measured on a tiny H3: backward growth was ~75 MB per block, now flat (61-88 MB at 2, 8 and 16 blocks).
The in-place mul_/addcmul_ on slices of the stream in the block helpers were the cause.
UNTESTED on the real weights.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01MpnF7GwNGMbngU11wjym7P

2026-10-17 19:47:34 +00:00

..

lora

H3 training: modulate blocks out of place so the backward pass stops keeping every block's recompute alive

2026-10-17 19:47:34 +00:00

__init__.py

LoRA training add-on: sets, tagger, trainer node on ComfyUI's Train LoRA, H3 support

2026-10-17 15:05:57 +00:00