Ltx2Guidance (src/vllm/model_executor/models/ltx2_pipeline.cpp:526) is the
seam "a caller reaches for when it holds a configured kind", per its own header
comment. Nothing in the product holds one.
Measured on row/LTX25-T2A-ONE-STAGE at c1fe35592:
$ git grep -n 'Ltx2Guidance(' -- src include examples tests
include/vllm/model_executor/models/ltx2_pipeline.h:330: the declaration
src/vllm/model_executor/models/ltx2_pipeline.cpp:526: the definition
tests/vllm/models/test_ltx2_pipeline.cpp:710: the only call
Two ported guiders are reachable through it and through nothing else:
| Symbol |
Reached from |
Ltx2CfgDelta |
Ltx2Guidance (ltx2_pipeline.cpp:532), test_ltx2_pipeline.cpp:615 |
Ltx2StgDelta |
Ltx2Guidance (ltx2_pipeline.cpp:534), test_ltx2_pipeline.cpp:630 |
Ltx2BatchedPerturbationConfig (ltx2_pipeline.h:380, four public members) is
the same shape from a different direction: it is constructed only at
test_ltx2_pipeline.cpp:832-859. The LTX-2.5 text-to-audio path perturbs through
Ltx2DitPerturbation, which is a different type, and no other path perturbs at
all.
This is the "test-only driver" shape .agents/reachability.md names: a unit test
that constructs the type by hand proves the class works, never that anything
reaches it. All four landed with #641 and predate every row since.
Ltx2MultiModalGuidance was the fourth member of this set and is no longer one:
#1005 gave it a production call site in
src/vllm/model_executor/models/ltx2_t2a.cpp.
What closing this means. Either route a product path through Ltx2Guidance
with a configured Ltx2GuiderKind — upstream's guider factory
(ltx-core components/guiders.py:294-342) is the behaviour it mirrors and
Ltx2GuiderSigmaBin / Ltx2GuiderParamsForSigma are already ported beside it —
or retire the unreached arms into .agents/completed/ with their provenance.
Deciding which is the work; this issue only records that the decision is owed.
Found while repairing the fresh review of #1039 on PR #1032. Out of scope there:
that row's spec §6b now states the true position and lists this issue under
## Owed.
Ltx2Guidance(src/vllm/model_executor/models/ltx2_pipeline.cpp:526) is theseam "a caller reaches for when it holds a configured kind", per its own header
comment. Nothing in the product holds one.
Measured on
row/LTX25-T2A-ONE-STAGEatc1fe35592:Two ported guiders are reachable through it and through nothing else:
Ltx2CfgDeltaLtx2Guidance(ltx2_pipeline.cpp:532),test_ltx2_pipeline.cpp:615Ltx2StgDeltaLtx2Guidance(ltx2_pipeline.cpp:534),test_ltx2_pipeline.cpp:630Ltx2BatchedPerturbationConfig(ltx2_pipeline.h:380, four public members) isthe same shape from a different direction: it is constructed only at
test_ltx2_pipeline.cpp:832-859. The LTX-2.5 text-to-audio path perturbs throughLtx2DitPerturbation, which is a different type, and no other path perturbs atall.
This is the "test-only driver" shape
.agents/reachability.mdnames: a unit testthat constructs the type by hand proves the class works, never that anything
reaches it. All four landed with #641 and predate every row since.
Ltx2MultiModalGuidancewas the fourth member of this set and is no longer one:#1005 gave it a production call site in
src/vllm/model_executor/models/ltx2_t2a.cpp.What closing this means. Either route a product path through
Ltx2Guidancewith a configured
Ltx2GuiderKind— upstream's guider factory(
ltx-core components/guiders.py:294-342) is the behaviour it mirrors andLtx2GuiderSigmaBin/Ltx2GuiderParamsForSigmaare already ported beside it —or retire the unreached arms into
.agents/completed/with their provenance.Deciding which is the work; this issue only records that the decision is owed.
Found while repairing the fresh review of #1039 on PR #1032. Out of scope there:
that row's spec §6b now states the true position and lists this issue under
## Owed.