Tools

Everything in tools/ is a standalone binary built against libllama plus common/. Each tool is a small program (typically one or two .cpp files) with its own CMakeLists.txt and README.md. Some tools also pull in extra ggml backends (e.g. llama-bench benchmarks every backend, llama-rpc-server ships its own networking).

This wiki uses the tools lens (matching the repo's own directory name) for these binaries; collectively they are the project's "applications". For the underlying library subsystems they exercise, see Systems.

Pages

• llama-cli — the primary chat / completion CLI.
• llama-server — OpenAI-compatible HTTP server with WebUI.
• llama-quantize — produce quantized GGUFs.
• llama-imatrix — produce importance-matrix files for IQ-quants.
• llama-perplexity — perplexity / KL-div / HellaSwag evaluator.
• llama-bench — throughput benchmarks.
• llama-mtmd-cli and clip — multimodal CLI plus the CLIP-based vision encoder.
• Other tools — gguf-split, gguf, batched-bench, rpc-server, tokenize, tts, completion, parser, fit-params, cvector-generator, export-lora, results.

Tool index

Binary	Source dir	Owner (CODEOWNERS)	Brief
llama-cli	tools/cli/	@ngxson	Interactive chat / single-shot generation
llama-server	tools/server/	@ggml-org/llama-server	OpenAI-compatible HTTP server + WebUI
llama-quantize	tools/quantize/	@ggerganov	Quantize GGUFs
llama-imatrix	tools/imatrix/	(no listed owner)	Importance matrix generator
llama-perplexity	tools/perplexity/	@ggerganov	Perplexity / quality evaluator
llama-bench	tools/llama-bench/	(no listed owner)	Throughput benchmark
llama-batched-bench	tools/batched-bench/	@ggerganov	Batched-decoding microbench
llama-mtmd-cli	tools/mtmd/	@ggml-org/llama-mtmd	Multimodal (vision/audio) CLI
llama-gguf-split	tools/gguf-split/	(no listed owner)	Split / merge GGUFs
rpc-server	tools/rpc/	@ggml-org/ggml-rpc	Remote ggml backend server
llama-tokenize	tools/tokenize/	@ggerganov	CLI tokenizer / detokenizer
llama-tts	tools/tts/	@ggerganov	Text-to-speech driver (uses an LLM + audio decoder)
llama-completion	tools/completion/	@ggerganov	FIM-style completion harness
llama-parser	tools/parser/	@pwilkin	Standalone PEG/autoparser CLI
llama-fit-params	tools/fit-params/	(no listed owner)	Curve-fits sampling/quant parameters
llama-cvector-generator	tools/cvector-generator/	(no listed owner)	Build control vectors
llama-export-lora	tools/export-lora/	(no listed owner)	Merge LoRA into a GGUF
tools/results/	—	(no listed owner)	Helper for storing/parsing benchmark results

The examples/ directory has additional smaller demos (Android, Swift, vim plugin, simple-chat, retrieval, ...). They are not "shipped tools" in the same sense, so they are not enumerated individually here, but the most useful ones are referenced where they relate to a tool.