3 Commits

Include renames

Author	SHA1	Message	Date
WildAi	900ceda564	fix: resolve review issues W1-W9; suite green (1107 passed)	2026-08-25 16:09:05 +03:00
WildAi	867a346552	fix(hrdit): Anima crashes + node-order independence ... G1: wrapper now mirrors the real ComfyUI optimized_attention signature (mask/attn_precision/skip_reshape/skip_output_reshape in slots 5-8) and forwards orig() with the correct positional order - the pre-fix wrapper fed skip_reshape into the mask slot on the unmasked Anima/cosmos path (AttributeError: bool.ndim). G2: pytest mock _sdpa flipped to the real ComfyUI signature (it matched the wrapper's inverted convention, hiding G1); conformance tripwire added (tests/test_orig_call_convention.py). G3: HAP decline-guards - non-square attention (cross-attn) and plan/model head-count mismatch decline to plain attention with one-time logs instead of crashing or computing wrong math. G4: _hrdit_carry_state() re-applies HRDiT patcher attrs after ModelPatcher.clone() in both apply functions (+ _hrdit_state_ref indirection), so SPA->HAP / HAP->SPA chaining is order-independent. SPA non-square guard: averaged passes apply spatial RoPE to both q and k (only valid for square self-attention); Anima cross-attention (q_len != k_len) crashed einsum - now declined to plain attention. README: v2.7.1 changelog + HAP model-specificity/v1-limitation notes.	2026-08-16 19:02:29 +03:00
WildAi	b9578e7400	feat(hrdit): add HAP node, calibration, scaling ... - HAP runtime: per-head scope plans via FlexAttention (CUDA + torch>=2.5) with SDPA dense-mask fallback - Calibration: Taylor-softmax scoring + knapsack solver (calibration/calibrate_hap.py, --dry_run toy pipeline) - Shipped FLUX scope plan (configs/scope_plan_flux.json) - SPA+HAP compose via refcounted shared hook install - proportional_attention knob on SPA+HAP (default off) - spa_layer_filter knob on SPA (flat index spec) - Demote "SPA averaged-attention ACTIVE" log to debug - README: HAP section, coverage matrix, v2.7.0 changelog	2026-08-16 01:41:20 +03:00