Fix stale seqlen cache in cascade reduce() (MPS crash, CUDA latent)

[Kyle-Wang0211]

Summary

VarLenTensor.reduce() in trellis2/modules/sparse/basic.py reads self.seqlen
from a per-scale _spatial_cache slot that, in the cascade Shape-SLat sampler's
CFG-rescale path (x_0_pos.std(...)), can hold a value left over from the parent
tensor's scale. When that happens, sum(seqlen) != feats.shape[0].

• MPS (Apple Silicon) — torch.segment_reduce has no Metal kernel and PyTorch falls back to CPU. The CPU path strictly enforces sum(lengths) == data.size(0) and raises RuntimeError. The cascade Shape-SLat pass aborts at ~step 3.
• CUDA — the existing call would not raise the same check in our reading. If the root-cause hypothesis below is correct, the CFG-rescale std() in cascade mode would return numerically wrong values silently. Easy to verify by adding assert lengths.sum() == data.size(0) before the existing call. We do not have CUDA hardware to confirm.

Closes the Apple-Silicon side of #74.

Repro (MPS)

git clone https://github.com/shivampkumar/trellis-mac
cd trellis-mac && bash setup.sh && source .venv/bin/activate
PYTORCH_ENABLE_MPS_FALLBACK=1 python generate.py path/to/image.png \
  --pipeline-type 1024_cascade --texture-size 2048 --steps 16

Aborts ~step 3 of the cascade Shape-SLat pass with:

File "trellis2/modules/sparse/basic.py", line 283, in reduce
    red = torch.segment_reduce(red, reduce=op, lengths=self.seqlen)
RuntimeError: segment_reduce(): Expected all rows of lengths along axis
to sum to data.size(lengths.dim()-1) when !unsafe.

Root cause hypothesis

1. SparseTensor.seqlen is cached in _spatial_cache[str(self._scale)]['seqlen'] (basic.py:543-548).
2. SparseTensor.replace() (basic.py:671-676) constructs the new tensor with spatial_cache=self._spatial_cache — same dict reference, same scale key. __merge_sparse_cache() (basic.py:705-715) merges across two sparse operands.
3. In the cascade Shape-SLat pass, flow_euler.sample_once derives x_0_pos from _pred_to_xstart(x_t, t, pred_pos), an elementwise chain over x_t and the model output. The result inherits x_t's _spatial_cache.
4. CFG-rescale (classifier_free_guidance_mixin.py:23) calls x_0_pos.std(dim=list(range(1, x_0_pos.ndim)), keepdim=True), which lands in VarLenTensor.reduce. self.seqlen returns the inherited cached value, which no longer matches x_0_pos.feats.shape[0] for this scale.

The base Shape-SLat pass does not exercise the same scale interaction and stays valid.

Fix

Two defensive layers in VarLenTensor.reduce():

1. Cache validation — if sum(lengths) != data.size(0), recompute lengths from coords (the authoritative source) and evict the stale seqlen / cum_seqlen / batch_boardcast_map / layout from the per-scale cache. On CUDA this branch is a no-op in the happy path; it only fires in the same stale-cache case that would silently miscompute today.
2. Explicit MPS → CPU fallback for segment_reduce — bypasses the auto fallback's noisier device-sync path. Also gives a ~5× speedup per cascade step (358s → ~50s) in our benchmark, independent of correctness.

Benchmark (M3 Pro 18GB, astronaut.jpeg, --pipeline-type 1024_cascade --texture-size 2048 --steps 16)

Stage	Before	After
Sparse structure (16 steps)	2:52	2:44
Shape SLat — base pass (16 steps)	1:15	0:54
Shape SLat — cascade pass (16 steps)	crash @ step 3	10:47 ✓
Texture SLat (16 steps)	n/a	6:17
Mesh + xatlas + Metal bake	n/a	0:25
End-to-end	crash	~24 min generation + 25s bake

Output: 21 MB GLB / 242 MB OBJ, 132K vert / 192K face, 2K PBR (base color + metallic + roughness + alpha).

Not verified

• CUDA regression on H100 / consumer cards — no hardware. The CUDA path is identical to before in the happy case; only behavior is in the stale-cache branch where it now uses a valid lengths instead of a stale one.
• 1536_cascade — only 1024_cascade was end-to-end-verified. The cache logic is scale-agnostic but the 1536 path has separate reports (see related).

Related cascade-pass reports

Repo	Issue	Symptom	Plausibly same root
visualbruno/ComfyUI-Trellis2	#113	1536_cascade abort, RTX 4090, spatial2channel.py:77 OOB	different file, possibly related fragility
visualbruno/ComfyUI-Trellis2	#98	1024_cascade / 1536_cascade, RTX 5090, max(): Expected reduction dim 0 to have non-zero size	empty-tensor downstream of stale cache?
visualbruno/ComfyUI-Trellis2	#31	cascade sparse_structure_resolution=64 edge case	unrelated, but cascade-path
pytorch/pytorch	#186540	native Metal segment_reduce kernel (open)	will obsolete layer 2 once merged

---

(感谢 TRELLIS.2 团队的开源——repro 在 macOS 上 100% 可复现，方便的话可以在 CUDA 上跑一下 cascade 输出比对，验证 silent miscompute 的影响范围。)

[microsoft-github-policy-service]

@Kyle-Wang0211 please read the following Contributor License Agreement(CLA). If you agree with the CLA, please reply with the following information.

@microsoft-github-policy-service agree [company="{your company}"]

Options:

• (default - no company specified) I have sole ownership of intellectual property rights to my Submissions and I am not making Submissions in the course of work for my employer.

@microsoft-github-policy-service agree

• (when company given) I am making Submissions in the course of work for my employer (or my employer has intellectual property rights in my Submissions by contract or applicable law). I have permission from my employer to make Submissions and enter into this Agreement on behalf of my employer. By signing below, the defined term “You” includes me and my employer.

@microsoft-github-policy-service agree company="Microsoft"

Contributor License Agreement

Contribution License Agreement

This Contribution License Agreement (“Agreement”) is agreed to by the party signing below (“You”),
and conveys certain license rights to Microsoft Corporation and its affiliates (“Microsoft”) for Your
contributions to Microsoft open source projects. This Agreement is effective as of the latest signature
date below.

1. Definitions.
   “Code” means the computer software code, whether in human-readable or machine-executable form,
   that is delivered by You to Microsoft under this Agreement.
   “Project” means any of the projects owned or managed by Microsoft and offered under a license
   approved by the Open Source Initiative (www.opensource.org).
   “Submit” is the act of uploading, submitting, transmitting, or distributing code or other content to any
   Project, including but not limited to communication on electronic mailing lists, source code control
   systems, and issue tracking systems that are managed by, or on behalf of, the Project for the purpose of
   discussing and improving that Project, but excluding communication that is conspicuously marked or
   otherwise designated in writing by You as “Not a Submission.”
   “Submission” means the Code and any other copyrightable material Submitted by You, including any
   associated comments and documentation.
2. Your Submission. You must agree to the terms of this Agreement before making a Submission to any
   Project. This Agreement covers any and all Submissions that You, now or in the future (except as
   described in Section 4 below), Submit to any Project.
3. Originality of Work. You represent that each of Your Submissions is entirely Your original work.
   Should You wish to Submit materials that are not Your original work, You may Submit them separately
   to the Project if You (a) retain all copyright and license information that was in the materials as You
   received them, (b) in the description accompanying Your Submission, include the phrase “Submission
   containing materials of a third party:” followed by the names of the third party and any licenses or other
   restrictions of which You are aware, and (c) follow any other instructions in the Project’s written
   guidelines concerning Submissions.
4. Your Employer. References to “employer” in this Agreement include Your employer or anyone else
   for whom You are acting in making Your Submission, e.g. as a contractor, vendor, or agent. If Your
   Submission is made in the course of Your work for an employer or Your employer has intellectual
   property rights in Your Submission by contract or applicable law, You must secure permission from Your
   employer to make the Submission before signing this Agreement. In that case, the term “You” in this
   Agreement will refer to You and the employer collectively. If You change employers in the future and
   desire to Submit additional Submissions for the new employer, then You agree to sign a new Agreement
   and secure permission from the new employer before Submitting those Submissions.
5. Licenses.

• Copyright License. You grant Microsoft, and those who receive the Submission directly or
  indirectly from Microsoft, a perpetual, worldwide, non-exclusive, royalty-free, irrevocable license in the
  Submission to reproduce, prepare derivative works of, publicly display, publicly perform, and distribute
  the Submission and such derivative works, and to sublicense any or all of the foregoing rights to third
  parties.
• Patent License. You grant Microsoft, and those who receive the Submission directly or
  indirectly from Microsoft, a perpetual, worldwide, non-exclusive, royalty-free, irrevocable license under
  Your patent claims that are necessarily infringed by the Submission or the combination of the
  Submission with the Project to which it was Submitted to make, have made, use, offer to sell, sell and
  import or otherwise dispose of the Submission alone or with the Project.
• Other Rights Reserved. Each party reserves all rights not expressly granted in this Agreement.
  No additional licenses or rights whatsoever (including, without limitation, any implied licenses) are
  granted by implication, exhaustion, estoppel or otherwise.

6. Representations and Warranties. You represent that You are legally entitled to grant the above
   licenses. You represent that each of Your Submissions is entirely Your original work (except as You may
   have disclosed under Section 3). You represent that You have secured permission from Your employer to
   make the Submission in cases where Your Submission is made in the course of Your work for Your
   employer or Your employer has intellectual property rights in Your Submission by contract or applicable
   law. If You are signing this Agreement on behalf of Your employer, You represent and warrant that You
   have the necessary authority to bind the listed employer to the obligations contained in this Agreement.
   You are not expected to provide support for Your Submission, unless You choose to do so. UNLESS
   REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING, AND EXCEPT FOR THE WARRANTIES
   EXPRESSLY STATED IN SECTIONS 3, 4, AND 6, THE SUBMISSION PROVIDED UNDER THIS AGREEMENT IS
   PROVIDED WITHOUT WARRANTY OF ANY KIND, INCLUDING, BUT NOT LIMITED TO, ANY WARRANTY OF
   NONINFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE.
7. Notice to Microsoft. You agree to notify Microsoft in writing of any facts or circumstances of which
   You later become aware that would make Your representations in this Agreement inaccurate in any
   respect.
8. Information about Submissions. You agree that contributions to Projects and information about
   contributions may be maintained indefinitely and disclosed publicly, including Your name and other
   information that You submit with Your Submission.
9. Governing Law/Jurisdiction. This Agreement is governed by the laws of the State of Washington, and
   the parties consent to exclusive jurisdiction and venue in the federal courts sitting in King County,
   Washington, unless no federal subject matter jurisdiction exists, in which case the parties consent to
   exclusive jurisdiction and venue in the Superior Court of King County, Washington. The parties waive all
   defenses of lack of personal jurisdiction and forum non-conveniens.
10. Entire Agreement/Assignment. This Agreement is the entire agreement between the parties, and
    supersedes any and all prior agreements, understandings or communications, written or oral, between
    the parties relating to the subject matter hereof. This Agreement may be assigned by Microsoft.