How to install llama.cpp with conda

The conda-forge package gives llama.cpp users prebuilt binaries inside an isolated conda environment instead of a source tree or host-wide install. It fits workstations and shared systems where the same shell may need several AI tools with different native libraries.

The package name in conda-forge is llama.cpp, while the installed command-line programs are llama-cli and llama-server. Keeping the install in a named environment makes the binary path explicit and keeps later model experiments separate from the base conda environment.

Use a conda-forge-only solve for this package so the compiler runtime and OpenMP libraries come from the same channel. Mixing default-channel dependencies into the environment can leave a binary that installs but cannot load its runtime libraries.

1. Open a terminal where conda is available.
2. Create a dedicated conda environment from conda-forge.

   $ conda create --yes --name llama-cpp --override-channels --channel conda-forge llama.cpp
   Channels:
    - conda-forge
   Platform: linux-aarch64
   Collecting package metadata (repodata.json): done
   Solving environment: done
   
   ## Package Plan ##
   
     environment location: /opt/conda/envs/llama-cpp
   
     added / updated specs:
       - llama.cpp
   
   The following NEW packages will be INSTALLED:
   
     libgomp            conda-forge/linux-aarch64::libgomp-15.2.0-h8acb6b2_19
     llama.cpp          conda-forge/linux-aarch64::llama.cpp-9851-cpu_openblas_h114b143_0
   ##### snipped #####
   #
   # To activate this environment, use
   #
   #     $ conda activate llama-cpp
   #

   The package index exposes the package as llama.cpp. The llama-cpp value is only the local environment name. The selected package build changes by platform and may be CPU, CUDA, Vulkan, or Metal.

3. Confirm that conda recorded the package in the new environment.

   $ conda list --name llama-cpp llama.cpp
   # packages in environment at /opt/conda/envs/llama-cpp:
   #
   # Name                     Version          Build                    Channel
   llama.cpp                  9851             cpu_openblas_h114b143_0  conda-forge

4. Verify that llama-cli starts from the environment.

   $ conda run --name llama-cpp llama-cli --version
   version: 1 (c14011d)
   built with GNU 14.3.0 for Linux aarch64

5. Verify that llama-server starts from the same environment.

   $ conda run --name llama-cpp llama-server --version
   version: 1 (c14011d)
   built with GNU 14.3.0 for Linux aarch64

   Use conda activate llama-cpp for an interactive shell, or keep conda run –name llama-cpp in scripts and automation. Add or download a GGUF model before running inference or starting the API server.
   Related: Run a local GGUF model with llama.cpp