nodes/olivv-cs-ComfyUI-FunPack
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

path-planner Phase 2d: conditioning-variant routing

Browse Source 
Route the conditioning variant from per-config outcome history, via the one
charter-safe surface: the INTERNAL value-function ascend+search (not user knobs,
so no WYSIWYG / hidden-settings violation). Sampler-owned visual ops stay on the
sampler (boundary law kept).

Directly serves the original complaint (Perfect not locking, Awful reproducing):
- _v2_path_conditioning_plan: matching-arm Beta posteriors -> liked-rate mean ->
  "lock" (>=0.6: hold the liked conditioning, skip the stochastic search),
  "explore" (<=0.4: widen the search to flee a disliked region), else "normal"
- value_function.search gains eps_scale to widen/tighten the search radius
- threaded conditioning_plan through ascend -> apply_scene_refinement_keys ->
  finalize; computed once per run from path_outcomes + prompt concepts; logged

Only active once the per-key VF is ready, since it rides the existing ascend path.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This commit is contained in:
digital-garbage
2026-06-21 16:26:16 +03:00
co-authored by Claude Opus 4.8
parent b6f17f2b3b
commit 9fb0ee267b
3 changed files with 85 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files
value_function.py
+7 -3
View File
@@ -117,14 +117,18 @@ class OnlineValueFunction(nn.Module):
prev_reward = reward_val
return c.detach().to(orig_device, orig_dtype)
def search(self, conditioning, n=16):
"""Monte Carlo conditioning search: score N random perturbations, return the best."""
def search(self, conditioning, n=16, eps_scale=1.0):
"""Monte Carlo conditioning search: score N random perturbations, return the best.
eps_scale widens (>1) or tightens (<1) the perturbation radius. The path planner uses
a wider radius to push a disliked config region toward a different conditioning, and the
caller skips this search entirely to LOCK a liked region's conditioning in place."""
if not self.is_ready():
return conditioning
mlp_device = next(self.parameters()).device
orig_device = conditioning.device
orig_dtype = conditioning.dtype
eps = conditioning.float().norm().item() * 0.025
eps = conditioning.float().norm().item() * 0.025 * float(eps_scale)
with torch.inference_mode(False), torch.no_grad():
c_base = torch.empty(conditioning.shape, dtype=torch.float32, device=mlp_device)
c_base.copy_(conditioning)