Llama Command Line Interface

This package provides a command line interface for the @hpcc-js/wasm-llama package. It exposes the bundled llama.cpp main function and can load GGUF models from the local drive or from HTTP(S) URLs, including Hugging Face model URLs.

To call wasm-llama-cli without installing:

npx @hpcc-js/wasm-llama-cli [options] [llama.cpp options]

To install the global command wasm-llama-cli via NPM:

npm install --global @hpcc-js/wasm-llama-cli

Usage:

Usage: wasm-llama-cli [options] [llama.cpp options]

Options:
  -m, --model <path-or-url>  Load a GGUF model from the local drive or an HTTP(S)
                             URL. Hugging Face /blob/ URLs are normalized to
                             /resolve/ URLs automatically.
      --llama-help           Show llama.cpp main help
      --version              Show bundled llama.cpp version
  -h, --help                 Show this help message

All other arguments are forwarded to llama.cpp main. If --model is supplied,
the model is loaded into the WASM filesystem and passed to main automatically.
The -- separator is optional; use it only when a llama.cpp argument must be
protected from wrapper option parsing.

Examples:

npx @hpcc-js/wasm-llama-cli --model https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct-GGUF/resolve/main/qwen2.5-0.5b-instruct-q4_k_m.gguf -p "Tell me a short friendly story about a tiny llama." -n 96
npx @hpcc-js/wasm-llama-cli -m ./model.gguf -p "Tell me a short friendly story about a tiny llama." -n 96
npx @hpcc-js/wasm-llama-cli --llama-help