No module named 'llama_cpp' in ComfyUI

If you searched for ModuleNotFoundError: No module named 'llama_cpp' or comfyui no module named llama_cpp, the important detail is the package name: the import is llama_cpp, but the package you usually install is llama-cpp-python.

This error often appears when a custom node for local LLM, VLM, QwenVL GGUF, prompt enhancement, or GGUF text inference tries to load llama_cpp during startup.

Symptoms

The startup log may show:

ModuleNotFoundError: No module named 'llama_cpp'

or:

Cannot import <path> module for custom nodes: No module named 'llama_cpp'

In Wonderful Launcher telemetry, this error repeatedly appeared with QwenVL GGUF style modules:

Error loading module AILab_QwenVL_GGUF_PromptEnhancer: No module named 'llama_cpp'

When that happens, ComfyUI itself may still open, but the custom node family that depends on llama_cpp will not register.

What llama_cpp Is

llama_cpp is the Python import module provided by llama-cpp-python, the Python bindings for llama.cpp.

In ComfyUI, custom nodes may use it to run local GGUF language or vision-language models. That is separate from ordinary Stable Diffusion image checkpoints and separate from many image-model GGUF loader nodes.

Common places this can appear:

Quick Diagnosis

For the official GitHub Windows portable package, run from the portable package root:

.\python_embeded\python.exe -s -c "import sys; print(sys.executable)"
.\python_embeded\python.exe -s -m pip show llama-cpp-python
.\python_embeded\python.exe -s -c "import llama_cpp; print('llama_cpp OK')"

For a manual Git install, activate the same venv that launches ComfyUI:

python -c "import sys; print(sys.executable)"
python -m pip show llama-cpp-python
python -c "import llama_cpp; print('llama_cpp OK')"

For ComfyUI Desktop or another managed launcher, use the app's environment tools. Do not assume the portable python_embeded folder exists.

Step 1: Decide Whether It Is Blocking

Fix llama_cpp when:

• the failing workflow uses QwenVL GGUF, VLM, local LLM, or prompt enhancer nodes
• the startup log says a custom node failed to import because of llama_cpp
• the workflow has red nodes from the same custom node family

Skip it for now when:

• ComfyUI opens and the workflow you need does not use local LLM/GGUF text nodes
• the visible issue is a missing image checkpoint, LoRA, VAE, or ControlNet model
• several unrelated plugins are failing at once and the earliest error is not llama_cpp

If many packages are failing together, start with ComfyUI Plugin Import Failed or ComfyUI Dependency Conflicts before adding another native package.

Step 2: Install the Package in the Right Python

For the official Windows portable package:

.\python_embeded\python.exe -s -m pip install llama-cpp-python

For a manual venv:

python -m pip install llama-cpp-python

That basic command may build from source. The official llama-cpp-python documentation also provides prebuilt wheel indexes for CPU and some CUDA builds. If a build fails, use a wheel that matches your Python and CUDA setup instead of repeatedly retrying the same command.

Step 3: Use CPU First Unless You Need GPU Acceleration

If your goal is only to make the custom node import, a CPU wheel can be the lowest-risk first repair:

.\python_embeded\python.exe -s -m pip install llama-cpp-python --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cpu

For a manual venv:

python -m pip install llama-cpp-python --extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cpu

Use a CUDA wheel only when the node actually runs local LLM inference and CPU performance is not enough. Match the wheel to your Python and CUDA version; do not paste a CUDA wheel URL from another machine.

Step 4: Verify the Import

Run:

.\python_embeded\python.exe -s -c "import llama_cpp; print(getattr(llama_cpp, '__file__', 'llama_cpp OK'))"

For a manual venv:

python -c "import llama_cpp; print(getattr(llama_cpp, '__file__', 'llama_cpp OK'))"

If this fails, ComfyUI will fail too. Fix the Python environment before restarting.

Common Failure Cases

What Not To Do

• Do not search for a package literally named llama_cpp first; the common package name is llama-cpp-python.
• Do not install into system Python if ComfyUI uses embedded Python or a venv.
• Do not install CUDA wheels blindly just because the machine has an NVIDIA GPU.
• Do not treat this as a normal image-model GGUF problem unless the traceback names llama_cpp.
• Do not reinstall all of ComfyUI before checking the custom node import failure.

How Wonderful Launcher Helps

This error is a good fit for environment-aware repair because the wrong Python is the most common trap. Wonderful Launcher helps by keeping installs separated, preserving the startup log, and making it easier to identify which environment actually starts ComfyUI before changing packages.

If llama_cpp is only one of many import failures, use the log to fix the earliest dependency first instead of adding more packages blindly.

Related Guides

• ComfyUI Plugin Import Failed
• ComfyUI-GGUF: Fix No Module Named 'gguf'
• ComfyUI Dependency Conflicts
• No Module Named 'onnx' or 'onnxruntime'
• ComfyUI No Module Named Error: When It Is Safe to Ignore

Source References

• llama-cpp-python GitHub repository
• llama-cpp-python PyPI package
• ComfyUI issue with No module named 'llama_cpp'
• ComfyUI Desktop issue with Qwen VL and llama-cpp-python wheel install