create llama-server command from hf cache list

llama-server-helper.sh

#!/bin/bash

# Ensure the script exits if a command fails unexpectedly
set -e

# Fetch and process HuggingFace cache list in a single pipeline
# awk filters for lines starting with 'model/', trims it, and prints the ID
mapfile -t models < <(hf cache list --no-truncate 2>&1 | awk '/^model\// {sub(/^model\//, ""); print $1}')

# Check if we actually found any models
if [ ${#models[@]} -eq 0 ]; then
    echo "No models found in cache (or 'hf cache list' failed)."
    exit 1
fi

echo "Available HuggingFace Models:"
echo "----------------------------"

# Use Bash's native 'select' to handle the menu UI and validation
PS3="Enter the number of the model to use (or type Ctrl+C to quit): "
select selected_model in "${models[@]}"; do
    if [ -n "$selected_model" ]; then
        break # Valid choice made, exit the menu loop
    else
        echo "Invalid selection. Please try again."
    fi
done

# Construct the initial command
llama_server_command="llama-server --host 0.0.0.0 --port 8080 -hf \"$selected_model\""

# Allow user to edit the command interactively (-e enables Readline, -i pre-fills text)
echo -e "\nReview or edit your execution command below:"
read -e -i "$llama_server_command" final_command

# Validate the final command structure before execution
if [[ ! "$final_command" =~ ^llama-server.*--host.*--port.*-hf ]]; then
    echo "Error: Invalid command format. Command must start with:"
    echo "  llama-server --host ... --port ... -hf ..."
    echo ""
    echo "Example allowed: llama-server --host 0.0.0.0 --port 8080 -hf <model-path>"
    exit 1
fi

# Check for dangerous patterns (optional additional security)
if [[ "$final_command" =~ (rm|cp|mv|wget|curl|nc|netcat|bash|sh\s+-i) ]]; then
    echo "Error: Dangerous command detected. Only llama-server execution allowed."
    exit 1
fi
echo "Running: $final_command"
bash -c "$final_command"