From 0.35.0: DLL load failed while importing libtriton: A dynamic link library (DLL) initialization routine failed.

[tin2tin]

Describe the bug

From 0.35.0, Triton seems to be required to run Diffusers, however no matter what version or whl I dig up, I end up with the above-mentioned error. Downgrading to 0.34.0, it works as it used to.

The error has stalled the development of my Diffusers based free add-on for Blender: Pallaidium. And users report the same error.
(Since it is using Blender's Python, the possible whls must be build for Python 3.11)

I've tried the windows-triton builds and any supposedly compatible triton whl on HF, I even tried to build my own whl. And none of it worked.

What do we do?

Reproduction

Running Diffusers 0.35.0 with any Windows Triton whl out there (...maybe?).

Logs

RuntimeError: Failed to import diffusers.models.transformers.transformer_flux because of the following error (look up to see its traceback):
DLL load failed while importing libtriton: A dynamic link library (DLL) initialization routine failed.
Error: Python: Traceback (most recent call last):
  File "...\python\Lib\site-packages\diffusers\utils\import_utils.py", line 953, in _get_module
    return importlib.import_module("." + module_name, self.__name__)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "...\python\Lib\importlib\__init__.py", line 126, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<frozen importlib._bootstrap>", line 1204, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1176, in _find_and_load
  File "<frozen importlib._bootstrap>", line 1126, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
  File "<frozen importlib._bootstrap>", line 1204, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1176, in _find_and_load
  File "<frozen importlib._bootstrap>", line 1147, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 690, in _load_unlocked
  File "<frozen importlib._bootstrap_external>", line 940, in exec_module
  File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
  File "...\python\Lib\site-packages\diffusers\models\transformers\__init__.py", line 20, in <module>
    from .transformer_chroma import ChromaTransformer2DModel
  File "...\python\Lib\site-packages\diffusers\models\transformers\transformer_chroma.py", line 33, in <module>
    from .transformer_flux import FluxAttention, FluxAttnProcessor
  File "...\python\Lib\site-packages\diffusers\models\transformers\transformer_flux.py", line 29, in <module>
    from ..attention_dispatch import dispatch_attention_fn
  File "...\python\Lib\site-packages\diffusers\models\attention_dispatch.py", line 72, in <module>
    from sageattention import (
  File "...\python\Lib\site-packages\sageattention\__init__.py", line 1, in <module>
    from .core import sageattn, sageattn_varlen
  File "...\python\Lib\site-packages\sageattention\core.py", line 20, in <module>
    from .triton.quant_per_block import per_block_int8 as per_block_int8_triton
  File "...\python\Lib\site-packages\sageattention\triton\quant_per_block.py", line 18, in <module>
    import triton
  File "...\python\Lib\site-packages\triton\__init__.py", line 20, in <module>
    from .runtime import (
  File "...\python\Lib\site-packages\triton\runtime\__init__.py", line 1, in <module>
    from .autotuner import (Autotuner, Config, Heuristics, autotune, heuristics)
  File "...\python\Lib\site-packages\triton\runtime\autotuner.py", line 9, in <module>
    from ..testing import do_bench, do_bench_cudagraph
  File "...\python\Lib\site-packages\triton\testing.py", line 7, in <module>
    from . import language as tl
  File "...\python\Lib\site-packages\triton\language\__init__.py", line 4, in <module>
    from . import math
  File "...\python\Lib\site-packages\triton\language\math.py", line 1, in <module>
    from . import core
  File "...\python\Lib\site-packages\triton\language\core.py", line 10, in <module>
    from ..runtime.jit import jit
  File "...\python\Lib\site-packages\triton\runtime\jit.py", line 12, in <module>
    from ..runtime.driver import driver
  File "...\python\Lib\site-packages\triton\runtime\driver.py", line 1, in <module>
    from ..backends import backends
  File "...\python\Lib\site-packages\triton\backends\__init__.py", line 50, in <module>
    backends = _discover_backends()
               ^^^^^^^^^^^^^^^^^^^^
  File "...\python\Lib\site-packages\triton\backends\__init__.py", line 43, in _discover_backends
    compiler = _load_module(name, os.path.join(root, name, 'compiler.py'))
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "...\python\Lib\site-packages\triton\backends\__init__.py", line 12, in _load_module
    spec.loader.exec_module(module)
  File "...\python\Lib\site-packages\triton\backends\amd\compiler.py", line 2, in <module>
    from triton._C.libtriton import ir, passes, llvm, amd
ImportError: DLL load failed while importing libtriton: A dynamic link library (DLL) initialization routine failed.

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "...\python\Lib\site-packages\diffusers\utils\import_utils.py", line 955, in _get_module
    raise RuntimeError(
RuntimeError: Failed to import diffusers.models.transformers.transformer_flux because of the following error (look up to see its traceback):
DLL load failed while importing libtriton: A dynamic link library (DLL) initialization routine failed.

System Info

accelerate: 1.11.0.dev0
aiofiles: 24.1.0
aiohappyeyeballs: 2.6.1
aiohttp: 3.12.15
aiosignal: 1.4.0
alias-free-torch: 0.0.6
annotated-types: 0.7.0
antlr4-python3-runtime: 4.9.3
anyio: 4.9.0
appdirs: 1.4.4
argbind: 0.3.9
asdff: 0.1.0
asttokens: 3.0.0
attrs: 25.3.0
audioread: 3.0.1
aura-sr: 0.0.4
auraloss: 0.4.0
autopep8: 2.3.1
av: 14.4.0
beautifulsoup4: 4.13.5
bitsandbytes: 0.47.0
blis: 1.2.1
boto3: 1.38.25
botocore: 1.38.25
braceexpand: 0.1.7
Brotli: 1.1.0
cached_path: 1.7.3
cachetools: 5.5.2
catalogue: 2.0.10
celluloid: 0.2.0
certifi: 2021.10.8
cffi: 1.17.1
cfgv: 3.4.0
chardet: 5.2.0
charset-normalizer: 2.0.10
chatterbox-tts: 0.1.1
clean-fid: 0.1.35
click: 8.2.1
clip-anytorch: 2.6.0
cloudpathlib: 0.21.1
cloudpickle: 3.1.1
cmudict: 1.0.32
colorama: 0.4.6
colorlog: 6.9.0
confection: 0.1.5
configparser: 7.2.0
conformer: 0.3.2
contourpy: 1.3.2
controlnet_aux: 0.0.10
cupy-cuda12x: 12.3.0
cycler: 0.12.1
cymem: 2.0.11
Cython: 3.0.11
da_core_news_md: 3.8.0
datasets: 4.0.0
dctorch: 0.1.2
decorator: 5.2.1
deepspeed: 0.16.5
descript-audio-codec: 1.0.0
descript-audio-codec-unofficial: 1.0.0
descript-audiotools: 0.7.2
descript-audiotools-unofficial: 0.7.4
diffusers: 0.35.0
dill: 0.3.8
distlib: 0.3.9
docker-pycreds: 0.4.0
docstring_parser: 0.16
einops: 0.8.1
einops-exts: 0.0.4
einx: 0.3.0
ema-pytorch: 0.7.7
en_core_web_md: 3.8.0
encodec: 0.1.1
executing: 2.2.0
f5-tts: 1.1.8
fastapi: 0.115.12
fastcore: 1.8.7
fastprogress: 1.0.3
fastrlock: 0.8.3
ffmpy: 0.5.0
filelock: 3.18.0
fire: 0.7.0
flatbuffers: 25.2.10
flatten-dict: 0.4.2
fonttools: 4.58.1
frozendict: 2.4.6
frozenlist: 1.7.0
fsspec: 2025.3.0
ftfy: 6.3.1
funcy: 2.0
future: 1.0.0
gguf: 0.17.1
gin-config: 0.5.0
gitdb: 4.0.12
GitPython: 3.1.44
google-api-core: 2.25.0rc1
google-auth: 2.40.2
google-cloud-core: 2.4.3
google-cloud-storage: 2.19.0
google-crc32c: 1.7.1
google-resumable-media: 2.7.2
googleapis-common-protos: 1.70.0
gradio: 5.39.0
gradio_client: 1.11.0
groovy: 0.1.2
grpcio: 1.71.0
h11: 0.16.0
h5py: 3.13.0
hf-xet: 1.1.5
hjson: 3.1.0
httpcore: 1.0.9
httpx: 0.28.1
huggingface-hub: 0.34.4
hydra-colorlog: 1.2.0
hydra-core: 1.3.2
HyperPyYAML: 1.2.2
identify: 2.6.12
idna: 3.3
image_gen_aux: 0.1.0.dev0
imageio: 2.37.0
imageio-ffmpeg: 0.6.0
imhist: 0.0.4
importlib_metadata: 8.7.0
importlib-resources: 5.12.0
imWatermark: 0.0.2
ipython: 9.5.0
ipython_pygments_lexers: 1.1.1
jax: 0.7.1
jaxlib: 0.7.1
jedi: 0.19.2
jieba: 0.42.1
Jinja2: 3.1.6
jmespath: 1.0.1
joblib: 1.5.1
jsonmerge: 1.9.2
jsonschema: 4.24.0
jsonschema-specifications: 2025.4.1
julius: 0.2.7
k-diffusion: 0.1.1
kiwisolver: 1.4.8
kornia: 0.8.1
kornia_rs: 0.1.9
laion_clap: 1.1.6
langcodes: 3.5.0
language_data: 1.3.0
lazy_loader: 0.4
librosa: 0.11.0
lightning-utilities: 0.14.3
llvmlite: 0.44.0
lmdb: 1.7.3
local-attention: 1.8.6
loguru: 0.7.3
marisa-trie: 1.2.1
Markdown: 3.8
markdown-it-py: 3.0.0
markdown2: 2.5.3
MarkupSafe: 3.0.2
matplotlib: 3.10.3
matplotlib-inline: 0.1.7
mdurl: 0.1.2
mediapipe: 0.10.21
meson: 0.63.0
ml_dtypes: 0.5.1
mmaudio: 1.0.0
mpmath: 1.3.0
msgpack: 1.1.0
multidict: 6.6.4
multiprocess: 0.70.16
murmurhash: 1.0.13
networkx: 3.4.2
ninja: 1.11.1.4
nitrous_ema: 0.0.1
nodeenv: 1.9.1
numba: 0.61.2
numpy: 1.26.4
omegaconf: 2.3.0
onnx: 1.18.0
open_clip_torch: 2.32.0
opencv-contrib-python: 4.11.0.86
opencv-python: 4.12.0.88
opencv-python-headless: 4.11.0.86
opt_einsum: 3.4.0
orjson: 3.10.18
packaging: 25.0
pandas: 2.0.2
parler_tts: 0.2.2
parso: 0.8.4
pathtools: 0.1.2
peft: 0.17.1
pillow: 11.3.0
pip: 25.2
platformdirs: 4.3.8
polars: 1.33.0
pooch: 1.8.2
pre_commit: 4.2.0
prefigure: 0.0.9
preshed: 3.0.10
progressbar: 2.5
prompt_toolkit: 3.0.51
propcache: 0.3.2
proto-plus: 1.26.1
protobuf: 3.20.1
psutil: 7.0.0
ptflops: 0.7.4
pure_eval: 0.2.3
py-cpuinfo: 9.0.0
pyarrow: 21.0.0
pyasn1: 0.6.1
pyasn1_modules: 0.4.2
pycodestyle: 2.12.1
pycparser: 2.22
pydantic: 2.10.6
pydantic_core: 2.27.2
pydub: 0.25.1
Pygments: 2.19.1
pyloudnorm: 0.1.1
pyparsing: 3.2.3
pyphen: 0.17.2
pypinyin: 0.54.0
pystoi: 0.4.1
python-dateutil: 2.9.0.post0
python-dotenv: 1.1.0
python-multipart: 0.0.20
pytorch-lightning: 2.1.0
pytz: 2025.2
PyWavelets: 1.4.1
PyYAML: 6.0.2
randomname: 0.2.1
referencing: 0.36.2
regex: 2024.11.6
reportlab: 4.4.1
requests: 2.32.5
resampy: 0.4.3
resemble-enhance: 0.0.2.dev241130080118
resemble-perth: 1.0.1
rich: 13.9.4
rpds-py: 0.25.1
rsa: 4.9.1
ruamel.yaml: 0.18.14
ruamel.yaml.clib: 0.2.12
ruff: 0.11.11
s3tokenizer: 0.2.0
s3transfer: 0.13.0
safehttpx: 0.1.6
safetensors: 0.6.2
sageattention: 2.1.1+cu124torch2.5.1
scikit-image: 0.25.2
scikit-learn: 1.2.2
scipy: 1.16.1
screenplain: 0.11.1
semantic-version: 2.10.0
sentencepiece: 0.1.99
sentry-sdk: 2.29.1
setproctitle: 1.3.6
setuptools: 63.2.0
shellingham: 1.5.4
six: 1.17.0
smart-open: 7.1.0
smmap: 5.0.2
sniffio: 1.3.1
sounddevice: 0.5.2
soundfile: 0.13.1
soupsieve: 2.7
soxr: 0.5.0.post1
spacy: 3.8.7
spacy-legacy: 3.0.12
spacy-loggers: 1.0.5
spandrel: 0.4.1
speechbrain: 0.5.16
srsly: 2.5.1
stable-audio-tools: 0.0.19
stack-data: 0.6.3
starlette: 0.46.2
suno-bark: 0.0.1a0
sympy: 1.13.1
tabulate: 0.9.0
tensorboard: 2.19.0
tensorboard-data-server: 0.7.2
tensordict: 0.8.3
termcolor: 3.1.0
textstat: 0.7.7
thinc: 8.3.4
threadpoolctl: 3.6.0
tifffile: 2025.5.26
timm: 1.0.20.dev0
tokenizers: 0.22.0
tomli: 2.2.1
tomlkit: 0.13.2
torch: 2.6.0+cu124
torch-stoi: 0.2.3
torchao: 0.12.0
torchaudio: 2.6.0+cu124
torchdiffeq: 0.2.5
torcheval: 0.0.7
torchlibrosa: 0.1.0
torchmetrics: 0.11.4
torchsde: 0.2.6
torchvision: 0.21.0+cu124
tqdm: 4.67.1
traitlets: 5.14.3
trampoline: 0.1.2
transformers: 4.57.0.dev0
transformers-stream-generator: 0.0.5
triton: 3.1.0
typer: 0.16.0
typing_extensions: 4.13.2
typing-inspection: 0.4.1
tzdata: 2025.2
ultralytics: 8.3.191
ultralytics-thop: 2.0.14
Unidecode: 1.4.0
urllib3: 1.26.8
uvicorn: 0.34.2
v-diffusion-pytorch: 0.0.2
vector-quantize-pytorch: 1.14.41
virtualenv: 20.31.2
vocos: 0.1.0
wandb: 0.15.4
wasabi: 1.1.3
wcwidth: 0.2.13
weasel: 0.4.1
webdataset: 0.2.100
websockets: 15.0.1
Werkzeug: 3.1.3
wget: 3.2
WhisperSpeech: 0.8
win32_setctime: 1.2.0
wrapt: 1.17.3
x-transformers: 2.3.9
xformers: 0.0.29.post3
xxhash: 3.5.0
yarl: 1.20.1
zipp: 3.22.0
zstandard: 0.16.0

Who can help?

@realsama @a-r-r-o-w

[tin2tin]

I wonder hasn't this happened to anyone else using 0.35.0+?
What version/build of Triton is compatible with the latest Diffusers on Windows?
Where can I find it?

[asomoza]

Hi, triton is not a requirement for diffusers, don't know why do you think that. Triton is only needed if you're using a function that requires it, like torch.compile. There's also the possibility that you installed another library that requires it.

I've been using diffusers from main, also tested 0.35 and 0.34, all of them worked without triton.

When I tested nunchaku, which requires triton and also diffusers, I installed the one from here and it worked fine without a problem if that helps you.

[tin2tin]

0.35.0 stops processing because of missing Triton, 0.34.0 doesn't. If may be another lib causing it, but it's being called differently in 0.35.0.

[DN6]

Based on the stack trace it looks like it could be an issue with sageattention that's installed in your environment. The new attention dispatch feature will attempt to import it if it's present in the environment. Could you try running in an environment without sageattention?

[tin2tin]

Tried without sageattention - still didn't work. This error from Diffusers writes: No module named 'triton

  ..\__init__.py", line 6636, in execute
    from diffusers import DiffusionPipeline, DPMSolverMultistepScheduler
  File "<frozen importlib._bootstrap>", line 1229, in _handle_fromlist
  File "..\python\Lib\site-packages\diffusers\utils\import_utils.py", line 965, in __getattr__
    value = getattr(module, name)
            ^^^^^^^^^^^^^^^^^^^^^
  File "..\python\Lib\site-packages\diffusers\utils\import_utils.py", line 964, in __getattr__
    module = self._get_module(self._class_to_module[name])
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "..\python\Lib\site-packages\diffusers\utils\import_utils.py", line 976, in _get_module
    raise RuntimeError(
RuntimeError: Failed to import diffusers.pipelines.pipeline_utils because of the following error (look up to see its traceback):
Failed to import diffusers.models.autoencoders.autoencoder_kl because of the following error (look up to see its traceback):
**No module named 'triton**'```

[asomoza]

mhmm this seems hard to debug on our side, there's a lot of libraries installed. Can you please share a requirements.txt file so I can test it. Also this happens with just running any pipeline code with diffusers?

This is not a diffusers only problem though, this is a mix of libraries that is causing the problem, but that gets triggered within diffusers because of triton.

Also, does it work on linux?

[tin2tin]

Isn't diffusers.models.autoencoders.autoencoder_kl needing triton? What libs are autoencoder_kl calling?

File "..\python\Lib\site-packages\diffusers\utils\import_utils.py", line 976, in _get_module
    raise RuntimeError(
RuntimeError: Failed to import diffusers.pipelines.pipeline_utils because of the following error (look up to see its traceback):
Failed to import diffusers.models.autoencoders.autoencoder_kl because of the following error (look up to see its traceback):
No module named 'triton'

I only use Windows. It does happen with sdxl and flux and flux kontext (haven't checked other options).

The add-on for Blender is called Palladium: https://github.com/tin2tin/Pallaidium It doesn't use requirements.txt - over the years there has been too much handholding of libs.

[asomoza]

Isn't diffusers.models.autoencoders.autoencoder_kl needing triton? What libs are autoencoder_kl calling?

As I told you, triton is not a direct dependency in diffusers if you don't use something that needs it.

I did a clean install of diffusers on windows and run a basic SDXL inference code and it worked, the installed libraries required for this are these:

accelerate==1.10.1
certifi==2025.8.3
charset-normalizer==3.4.3
colorama==0.4.6
diffusers==0.35.1
filelock==3.13.1
fsspec==2024.6.1
huggingface-hub==0.34.4
idna==3.10
importlib_metadata==8.7.0
Jinja2==3.1.4
MarkupSafe==2.1.5
mpmath==1.3.0
networkx==3.3
numpy==2.1.2
packaging==25.0
pillow==11.0.0
psutil==7.0.0
PyYAML==6.0.2
regex==2025.9.1
requests==2.32.5
safetensors==0.6.2
setuptools==70.2.0
sympy==1.13.3
tokenizers==0.22.0
torch==2.8.0+cu128
torchvision==0.23.0+cu128
tqdm==4.67.1
transformers==4.56.0
typing_extensions==4.12.2
urllib3==2.5.0
zipp==3.23.0

yours are like 5x more, and as you can see, I don't have triton installed.

I can also do this without a problem:

from diffusers.models.autoencoders.autoencoder_kl import AutoencoderKL

Sadly, even if I use and have blender (on linux) I don't have the time to test this on your add-on, you will need to give us a minimal reproducible code and environment with just diffusers.

I'll try to test it with windows with your addon but right now I don't have the spare time to do it.

edit: since your log had sage attention in it, I tried with installing triton-windows and Sage Attention from here matching my cuda version and it also worked, so that also doesn't seem to be the problem.

sageattention @ file:///C:/projects/python/diffusers/sageattention-2.2.0%2Bcu128torch2.8.0.post2-cp39-abi3-win_amd64.whl#sha256=b4540f844bfbbc35b9404953f2c11b16a06454a1ef1409975f3b3c8911e4417d
triton-windows==3.4.0.post20

[tin2tin]

Thanks for investing your time in helping me out. The add-on cover all types of AI generated media, so yes, that's a lot of dependencies. I guess, I'll have to remove all dependencies added to the project over the past half year, one by one, and see if it suddenly works.