Back to the README.
Llama GUI automatically creates a timestamped llama-gui-*.log file in the
checkout's logs/ folder on every launch. No debug option or special launcher is
needed. The log path is printed in the server terminal. After a crash, keep the
file from that run; restarting creates a new file rather than overwriting it.
The 10 newest session logs are retained, with older ones removed at startup when
possible. Files are not rotated during a running session so native crash reporting
can keep using the same open file descriptor.
Logs include UTC timestamps, Python/runtime details, backend stderr messages,
uncaught Python exceptions (including background threads), and native crash
tracebacks when Python's fault handler can report them. Normal server shutdowns
record the exit code. Existing console error output is preserved. This also works
for silent launches, GUI restarts, and Pinokio's server.py entrypoint.
These are GUI-server diagnostics; llama.cpp subprocess output remains in Monitor. Logs are local files and may contain paths and error details, so review them before sharing. If the log directory is unwritable, the server warns in its terminal and continues without file logging. A forced process kill, power loss, or hang may leave no final traceback; absence of a shutdown entry alone does not identify the cause.
App does not start at http://127.0.0.1:5240, or server launch fails on a taken port. Close the conflicting app or change its port.
Place .gguf files in the active models folder, refresh the model list in Configure, reset an unavailable custom folder, or use -hf / HF repo flags for remote loading.
Immediate crash or DLL/backend errors: reinstall a backend that matches your hardware/drivers, try Install → Repair Install, or test with CPU first.
On Linux, Llama GUI uses ldd when available to check llama-server, llama-cli, and packaged ggml backend plugins before launch, then reports unresolved shared libraries in the Install tab. If a repaired Vulkan or ROCm install still fails, verify the host driver stack directly:
ldd llama/bin/llama-server | grep "not found"
vulkaninfo --summary # Vulkan
rocminfo # ROCm / AMD kernel-driver accessIf the binaries run manually but ldd reports missing libraries (for example on
NixOS), close the GUI and start it from the checkout with the Linux-only opt-out:
LLAMA_GUI_SKIP_LDD=1 ./mac_linux_start.shThis skips the GUI's executable and ggml-plugin dependency probes for status, custom-backend activation, preflight, and launch. Missing executables, permissions, and actual runtime failures still apply; it does not supply missing libraries. Restart without the variable to restore the checks.
Lemonade ROCm archives include user-space ROCm libraries, but the selected gfx target must match the GPU and the host still needs working AMD kernel-driver access. If model loading runs unusually long, the app keeps the process stoppable and adds a persistent warning directing you to the live process output.
Install looks fine but binaries are missing: check quarantine, restore blocked llama/ files, and only add a project exclusion if you trust the source.
Upstream Windows x64 releases switched from CUDA 13.3 to 13.4 in b10978. In Install, select CUDA 13.4 (NVIDIA), choose a release, and click Install. Existing CUDA 13.3 configurations remain supported for older builds; the backend update button stays within the installed CUDA version. Both the binary and matching CUDA runtime archive must be available before a release can be installed.
The Install page discovers official llama.cpp CUDA/ROCm versions for your platform and architecture automatically. Use the refresh button beside Version to refresh both releases and backend options. New options do not change your installed backend; select a version and click Install to switch. Official ROCm packages still need their separate runtime, and discovery does not check your GPU/driver compatibility.
If GitHub is unavailable or rate-limited, saved options and built-in backends remain
available. The message below Backend reports discovery failures. Automatic checks
are cached for an hour; manual retries have a 30-second cooldown. Previously discovered
options survive restarts in llama/official-backends.json, although downloading a
release still requires a connection. Only complete CUDA binary/runtime pairs are
offered in the release list. Lemonade options keep their existing behavior.
Saved options can remain after upstream removes a package or its releases fall
outside the recent-release scan, leaving an empty Version list. Discovery only
recognizes supported package naming conventions; a new archive format or tag
convention may require a GUI update even when the toolkit version is numeric.
Need git on PATH and a git clone (not a zip extract). Retry from Install and read the update status text.
Rerun the platform install script so ddgs is present; check internet access; try a simpler query (free providers rate-limit). Leave Web Search off for offline chat.
Copy recent errors from the live process output on the Monitor tab. Retry a minimal setup (CPU, one local model, defaults). Include logs, backend, and model name when reporting issues.