digital-garbageandClaude Opus 5
8e3c707c81
The Chain Sampler reads one conditioning entry per SCENE (scene_conditionings = positive[:max_scenes]). When Studio stopped discarding the entries past the first of a wired CONDITIONING — the fix for a reference surviving while the prompt did not — those entries reached the sampler untagged, and the r2v reference block was counted as a scene of its own. The run sampled the shot twice, which from outside is "the model ran a second pass for no reason". Studio tags every pass-through entry as a companion; the sampler excludes companions when counting scenes and hands them to every scene instead, so they still reach the model. Both halves of the original bug stay fixed: nothing is dropped, and nothing is mistaken for a scene. Also fixes the liked-conditioning blend, which was the last payload path still doing a raw lerp: _v2_shape_compatible ignores sequence length by design, so a direction learned at 586 positions met a 494-position prompt, threw, and took the whole pull toward what the user rated well with it. Routed through _v2_payload_like like every other payload. Adds stage timers for the nine learning steps the first pass did not cover — 55% of a 26.9s run was accounted for, and the rest was unnamed. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
64 lines
2.3 KiB
Python
64 lines
2.3 KiB
Python
"""A wired r2v conditioning is several entries describing ONE generation, not several scenes.
|
|
|
|
The Chain Sampler reads one conditioning entry per SCENE. When Studio stopped discarding the
|
|
entries past the first of a wired CONDITIONING — the fix for "the right face speaking invented
|
|
syllables" — those entries started arriving at the sampler untagged, and it counted the
|
|
reference block as a second scene. From outside that looks like the model sampling the shot
|
|
twice for no reason anyone asked for.
|
|
"""
|
|
import sys
|
|
|
|
import pytest
|
|
import torch
|
|
|
|
sys.path.insert(0, ".")
|
|
|
|
|
|
@pytest.fixture
|
|
def chain():
|
|
from samplers import FunPackLTXAVSceneChainSampler
|
|
return FunPackLTXAVSceneChainSampler
|
|
|
|
|
|
def _entry(companion=False, text="scene"):
|
|
meta = {"funpack_scene_text": text}
|
|
if companion:
|
|
meta["funpack_companion_conditioning"] = True
|
|
return [torch.zeros(1, 4, 8), meta]
|
|
|
|
|
|
def test_a_tagged_entry_is_a_companion(chain):
|
|
assert chain._is_companion_conditioning(_entry(companion=True)) is True
|
|
|
|
|
|
def test_an_ordinary_entry_is_a_scene(chain):
|
|
assert chain._is_companion_conditioning(_entry()) is False
|
|
|
|
|
|
@pytest.mark.parametrize("junk", [None, [], [torch.zeros(1)], "text", 7, [torch.zeros(1), None]])
|
|
def test_a_malformed_entry_is_not_mistaken_for_a_companion(chain, junk):
|
|
"""Guessing wrong here drops a real scene, so anything unreadable counts as a scene."""
|
|
assert chain._is_companion_conditioning(junk) is False
|
|
|
|
|
|
def test_companions_do_not_add_scenes(chain):
|
|
"""The regression itself: one scene plus one reference companion is ONE scene."""
|
|
positive = [_entry(), _entry(companion=True)]
|
|
scenes = [c for c in positive if not chain._is_companion_conditioning(c)]
|
|
assert len(scenes) == 1
|
|
|
|
|
|
def test_real_scenes_still_count(chain):
|
|
positive = [_entry(text="a"), _entry(text="b"), _entry(companion=True)]
|
|
scenes = [c for c in positive if not chain._is_companion_conditioning(c)]
|
|
assert len(scenes) == 2
|
|
|
|
|
|
def test_studio_tags_what_it_passes_through():
|
|
"""Studio is what applies the tag; without it the sampler cannot tell the two apart."""
|
|
import inspect
|
|
|
|
from conditioning import FunPackVideoRefinerV2
|
|
src = inspect.getsource(FunPackVideoRefinerV2.refine_v2)
|
|
assert 'meta_copy["funpack_companion_conditioning"] = True' in src
|