[Qwen-Image 2.1 / Z-Image] Complex-valued RoPE hard-fails on backends without complex64 kernels (Ascend NPU: aclnnIndex EZ1001)
Maintainer antworten meist innerhalb von 1 Tag
Dieses Issue hat noch niemand übernommen.
Bewertung
- Schwierigkeit
- 4/5
- Geschätzter Aufwand
- 3-5 Tage
- Anfängerfreundlichkeit
- 52/100
- Issue-Typ
- Bug
- Klarheit
- Klar beschrieben
- Aktivitätsstatus
- Aktiv
- Bereich
- machine-learning
Rechercherichtung
Read transformer_qwenimage21.py, transformer_z_image.py, and the existing ROPE_PER_DEVICE and real-valued rotary path in transformer_qwenimage.py. Check the behavior described for Ascend NPU, CoreML export, and torch.compile, then trace how device-specific frequency tables are selected. Done means the three pipeline families use the requested real-valued path where needed and Qwen-Image 1.0 registers NPU support; no test file is named in the issue.
Vom Indexierungsmodell aus dem Issue-Text verfasst.
Beschreibung
Describe the bug
Two pipeline families in the Qwen-Image lineage still build their RoPE table with torch.polar and then index/multiply it as complex64 on the device, with no real-valued fallback:
transformer_qwenimage21.py—QwenImage21Rope.forwardhardcodesuse_real=Falsetransformer_z_image.py—torch.polar+view_as_complex
transformer_qwenimage.py (Qwen-Image 1.0) already has ROPE_PER_DEVICE from the AWS Neuron work in #13718, but only "neuron" is registered, so other accelerators keep taking the complex path.
On backends without complex64 kernels this is a hard failure rather than a slowdown, and it affects three targets, not one:
- Ascend NPU (torch_npu) — the op-plugin has no complex64
index/mul.QwenImage21Rope.forwardraisesaclnnIndex EZ1001as soon as the complex frequency table is indexed on device. Qwen-Image 2.1 and Z-Image currently cannot run on Ascend NPUs without an out-of-tree monkey-patch. - CoreML export —
torch.polarhas no MIL equivalent (apple/coremltools#2258). torch.compile— inductor cannot codegen complex ops (#12668), which matches this same code path.
CUDA eager is unaffected, which is presumably why this has stayed low-pressure upstream.
I understand #12668 already tracks the complex-op problem from the torch.compile angle; opening this as a separate issue because the backend failure mode and the fix scope are different (see "Describe the solution you'd like"). Happy to close this if you would rather consolidate into #12668.
Reproduction
The failure is device-specific, but the code path is easy to isolate. On an Ascend NPU with a from-source diffusers checkout:
import torch
from diffusers.models.transformers.transformer_qwenimage21 import QwenImage21Rope
# Any small instantiation of the rope module is enough; the failure happens at the
# first on-device indexing of the complex frequency table.
rope = QwenImage21Rope(...).to("npu") # or .to(torch.device("npu"))
rope(x, freqs_cis) # -> aclnnIndex EZ1001
A backend-agnostic sanity check for the same assumption (runs anywhere, no model weights needed):
import torch
freqs = torch.polar(torch.ones(8, 4), torch.rand(8, 4)).to("npu") # complex64 on device
freqs[torch.arange(8).to("npu")] # index -> unsupported on NPU
Describe the solution you'd like
Extend the existing #13718 pattern instead of adding a new one:
- port
ROPE_PER_DEVICE+ the real-valuedapply_rotary_emb_qwen_neuronpath intotransformer_qwenimage21.pyandtransformer_z_image.py; - keep the frequency table and its indexing on CPU, crossing the device boundary as reals (
_get_device_freqsalready does this for Neuron); - register
"npu"inROPE_PER_DEVICEfor Qwen-Image 1.0.
Context
I hit this while bringing up image models on a single Ascend 910B2; Qwen-Image 2.1 was the only pipeline in the set that could not run at all without patching.
Disclosure, and a note on scope: I opened PR #14905 for exactly this in September and it was auto-closed because it did not link an issue — that is on me, not on the change. Before re-opening anything I would like an explicit ack here, as the AI-assisted contribution section asks. I am also happy to split the work into one PR per pipeline instead of one size/L PR covering both, if that is easier to review.
Logs
RuntimeError: The inner error is reported as above. The process exits for this inner error. (Inner error: EZ1001: ... aclnnIndex ...)
File "diffusers/models/transformers/transformer_qwenimage21.py", line ..., in forward
...
(The exact op-level message from aclnnIndex is the one that names the unsupported complex64 dtype.)
System Info
- diffusers: main (from-source checkout at the time of report; PR #14905 was opened 2026-09-29)
- platform: Linux aarch64
- Python: 3.12.13
- torch: 2.10.0+cpu
- torch_npu: 2.10.0.post4
- CANN: 9.1.0
- accelerator: 1x Ascend 910B2 (HBM 64 GB)
Who can help?
Anyone with familiarity with the Qwen-Image RoPE implementations, or with Apple/Neuron backends where the same real-valued path already exists.
- Vorherrschende Sprache
- Python
- Sterne
- 34.9k
- Forks
- 7.5k
- Ø Merge
- 3 T. 16 Std.
- Gemergte PRs (30 T.)
- 65
Entwicklungsumgebung
- Kein Dockerfile und keine Docker-Compose-Datei
- Hat eine Pull-Request-Vorlage
- Beitragsleitfaden lesen
Erste Schritte
- Lesen Sie das ganze Issue und danach den Beitragsleitfaden des Projekts.
- Schreiben Sie ins Issue, dass Sie es übernehmen — das erspart doppelte Arbeit.
- Forken Sie das Repository und arbeiten Sie in einem Branch.
- Öffnen Sie einen Pull Request, der die Issue-Nummer nennt.
Mehr aus huggingface/diffusers
-
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 84/100
huggingface/diffusers#15363 ·
Maintainer antworten meist innerhalb von 1 Tag
-
`ZImageImg2ImgPipeline` / `ZImageInpaintPipeline`: explicit height/width different from the input image raises RuntimeErrorEvtl. vergeben @chessing234 hat das vor 1 Tag übernommen. Offenbug
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 70/100
huggingface/diffusers#15337 · 2 Kommentare ·
Maintainer antworten meist innerhalb von 1 Tag
-
Multi-vector textual inversion uses only the first vector when the token has uppercase letters (transformers v5)Evtl. vergeben @sunhouy hat das vor 2 Tagen übernommen. Offen
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 75/100
huggingface/diffusers#15313 ·
Maintainer antworten meist innerhalb von 1 Tag
-
TCDScheduler advances the generator on the final step of a strength-truncated custom scheduleEvtl. vergeben @SIDDARTHAREDDY8 hat das vor 5 Tagen übernommen. Offenbug
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 74/100
huggingface/diffusers#15288 · 2 Kommentare ·
Maintainer antworten meist innerhalb von 1 Tag
-
unload_textual_inversion(token) leaves the extra tokens of a multi-vector embedding behind, so it can't be loaded againEvtl. vergeben @sohailshk hat das vor 11 Tagen übernommen. Offen
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 72/100
huggingface/diffusers#15211 · 3 Kommentare ·
Maintainer antworten meist innerhalb von 1 Tag
Alle Issues in huggingface/diffusers
Ähnliche Issues
-
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 73/100
Maintainer antworten meist innerhalb von 1 Tag
-
Unused loot bag group names left in dreamcraft en_US.langEvtl. vergeben Ein verknüpfter Pull Request ist offen oder bereits gemergt. OffenStatus: Triage Suggestion
Schwierigkeit 1/5 Unter einer Stunde Anfängerfreundlichkeit 90/100
GTNewHorizons/GT-New-Horizons-Modpack#27789 ·
Maintainer antworten meist innerhalb von 1 Tag
-
bug frontend sev-5
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 83/100
-
core good first issue
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 65/100
-
b4b code health next
Schwierigkeit 2/5 1-3 Stunden Anfängerfreundlichkeit 68/100
ESCOMP/CTSM#4330 · 1 Kommentar ·
Maintainer antworten meist innerhalb von 1 Tag