xFormers is installed but unavailable: inspect its native extension and operators

Symptoms and scope

pip can show xFormers as installed while its C++/CUDA extension fails to load or no attention operator supports the input. Package presence and usable GPU kernels are separate checks.

Error fragments for search; IDs, values and filenames may vary:

xFormers can't load C++/CUDA extensions
No operator found

Source-supported context

xFormers supplies an info command and platform/Torch installation requirements. ComfyUI's troubleshooting separates core execution from third-party extensions. Source 1 Source 2 Source 3 ComfyUI startup options

Cases to distinguish

1. Installation targeted system Python rather than ComfyUI's runtime.

2. The binary targets a different Python/Torch/CUDA combination, or no operator supports the device, dtype or input layout.

3. A node requires xFormers even though the core could use another attention implementation.

Diagnostic sequence

The sequence below is an editorial procedure based on the cited context, not a diagnosis already confirmed for your environment.

Step 1. Run the info check in the actual interpreter and save operator availability and load errors, not just the successful pip output.

Step 2. Compare xFormers' current installation instructions with Torch, compute backend and hardware. Check whether installing the extension would replace working Torch dependencies.

Step 3. In an environment copy, use this ComfyUI version's --help to select a supported non-xFormers attention path, then compare the same native workflow. Do not assume the default path excludes xFormers or that one flag works in every version.

Step 4. Build from source only with confirmed support and prerequisites; otherwise use a compatible combination or a workflow that does not require the extension.

Commands and request examples

Use the environment and shell identified above. Paths and variables are examples to adapt; these commands/snippets were not executed for your machine. Portable examples run from the package root. A virtual environment or managed launcher can use a different executable layout.

.\python_embeded\python.exe -m xformers.info

Completion check

The relevant operator is listed and a small real request completes with the expected backend selected in the log.

Limitations and cautions

Avoid unverified repair wheels, which execute local code. Changing attention may change numerical results and performance.

Original sources

• xFormers README — original record checked 2026-09-21.
• PyTorch local installation — original record checked 2026-09-21.
• ComfyUI custom node troubleshooting — original record checked 2026-09-21.
• ComfyUI startup options source — checked 2026-09-25.

Localization prepared: 2026-09-22. The source-check dates above were inherited from the existing article; this translation does not claim they were all rechecked today. No GPU run, minimum-memory measurement or runtime guarantee is supplied.

Related troubleshooting and guides

• SageAttention or Triton import failure: match the platform and build combination
• No kernel image is available: check GPU architecture support in each binary