llama.sh

No-messing-around sh client for llama.cpp's server

NOTE: Default config optimized for Llama-3.1

Depends on

• sh
• curl
• jq

Setup

Add these scripts to your $PATH. Configuration is done by editing the top of the llama.sh file.

Not sure how to set up llama.cpp? Check out Mozilla's llamafile!

Usage

        llama.sh [--options] "prompt"
        echo "prompt" | llama.sh [--options]
        echo "prompt" | llama.sh [--options] "system prompt"
Flags:
        --n-predict N,   -n N        (number of tokens to generate, -1 for inf. default: -1)
        --temp TEMP,     -t TEMP     (temperature. default: 0.6)
        --min-p P,       -m P        (min-p. default: 0.05)
        --top-p P,       -p P        (top-p. default: 0.95)
        --top-k K,       -k K        (top-k. default: 45)
        --stop "word",   -s "word"   (stop word. default: none)
        --log logfile,   -l logfile  (set file for logging. default: ~/.cache/last_response.txt)
        --verbose,       -v          (echo json payload before sending)
        --raw,           -r          (do not wrap prompt with prefix/suffix strings)
        --no-sys-prompt, -e          (do not include system prompt string)
        --api-key "key", -a "key"    (override key used for llama.cpp API, usually not needed unless explicitly set. will override env var)
        --api-url "url", -u "url"    (override url used for llama.cpp API. will override env var)
        --help,          -h          (display this message)
Environment Variables:
        LSH_SYSTEM_PROMPT_PREFIX   (string prefixed to system prompt input)
        LSH_PREFIX                 (string prefixed to user prompt input)
        LSH_SUFFIX                 (string appended to user prompt input)
        LSH_API_KEY                (optional API key)
        LSH_API_URL                (API url)

Demo (python_chat.sh)