llama.cpp leverages the ggml tensor library in order to run
large language models (LLMs) provided in the GGUF file format.
+# We only distribute a few very useful tools, with stable CLI options
Package: llama-cpp-tools
Architecture: any
Depends: ${misc:Depends}, ${shlibs:Depends},
libllama0, ggml, curl
Description: Inference of large language models in pure C/C++ (tools)
- llama-cli: utility tool wrapping most features provided by libllama.
+ llama-cli: versatile tool wrapping most features provided by libllama.
It typically allows one to run one-shot prompts or to "chat"
with a large language model.
.
- llama-quantize: utility tool to "quantize" a large language model
+ llama-quantize: utility to "quantize" a large language model
GGUF file. Quantizing is the process of reducing the precision of
the underlying neural-network at aminimal cost to its accuracy.
.
+# Most executables produced are not stable enough to be distributed
/usr/bin/llama-*
/usr/libexec/*/ggml/llama-*
+# Python is not supported
+/usr/bin/*.py
+
+# Test executables are not distributed
/usr/bin/test-*
-/usr/bin/*.py
overview / pkg / ggml / sources / llama.cpp / commitdiff