llama_cpp failures: distinguish installation, native builds and model loading

Identify the failing stage

Upstream documents llama-cpp-python as the distribution behind llama_cpp, with source builds, prebuilt wheels and hardware-specific backends.1 A failed build is not the same as a missing model.

python -m pip show llama-cpp-python

Use the original ComfyUI interpreter. For an installation failure, retain the first concrete compiler or CMake error, not only the final wheel-failure line. Match the node's supported release, OS, Python architecture and intended backend before choosing an upstream build route.

After the import works

For a model-initialization error, record the loader, model origin and required companion files. Our suggested triage does not assume every .gguf file belongs to the same loader. Likewise, PyTorch's device report is not evidence of the backend this library used.

Test a small supported input and record installation, import, model initialization and inference separately. Try build-option changes in an environment copy. Do not reinstall the entire application to investigate a failure that only begins when one model is opened.

Related checks

• Nunchaku still fails after installation: align the node, wheel and model
• Where to run ComfyUI commands: PowerShell, CMD and virtual environments

Research draft. No local/GPU reproduction or independent editorial sign-off. Upstream statements and our recommendations are distinguished.