#!/bin/sh # Usage: ./test_prompts.sh [prompt_file] SOCK_PATH="/var/run/lexd/sock" EXECUTABLE="./lex" # Mandatory Model Argument MODEL_PATH="$1" if [ -z "$MODEL_PATH" ]; then echo "Usage: $0 [prompt_file]" >&2 exit 1 fi if [ ! -f "$MODEL_PATH" ]; then echo "Error: Model file '$MODEL_PATH' not found." >&2 exit 1 fi # Derive base model name and output result filename MODEL_NAME=$(basename "$MODEL_PATH") RESULT_FILE="results_${MODEL_NAME}.txt" # Default prompt filename (optional 2nd argument) PROMPT_FILE="${2:-test_prompts.txt}" # Ensure executable exists if [ ! -x "$EXECUTABLE" ]; then echo "Error: Executable '$EXECUTABLE' not found or not executable." >&2 exit 1 fi # Ensure prompt file exists if [ ! -f "$PROMPT_FILE" ]; then echo "Error: Prompt file '$PROMPT_FILE' not found." >&2 exit 1 fi # Helper function to get current swap occupancy in MB by parsing swapctl -l # OpenBSD swapctl -l outputs 512-byte block counts by default get_swap_occupancy_mb() { swapctl -l | awk ' NR > 1 { # Column 3 is Used blocks (512-byte blocks) used_blocks += $3 } END { # Convert 512-byte blocks to MB (blocks * 512 / 1024 / 1024) printf "%.2f", used_blocks * 512 / 1048576 }' } # Count valid prompts using POSIX awk total_prompts=$(awk '/^[[:space:]]*($|#)/ {next} {count++} END {print count+0}' "$PROMPT_FILE") if [ "$total_prompts" -eq 0 ]; then echo "No valid prompts found in '$PROMPT_FILE'." exit 0 fi # Create (or truncate) result file > "$RESULT_FILE" # Create horizontal rule RULE="" i=0 while [ $i -lt 72 ]; do RULE="${RULE}-" i=$((i + 1)) done # --- Start daemon --- echo "Starting daemon with model: '$MODEL_PATH'" "$EXECUTABLE" -m "$MODEL_PATH" & DAEMON_PID=$! cleanup() { kill "$DAEMON_PID" 2>/dev/null wait "$DAEMON_PID" 2>/dev/null rm -f "$TIME_LOG" } trap cleanup EXIT INT TERM waited=0 while [ ! -S "$SOCK_PATH" ]; do if ! kill -0 "$DAEMON_PID" 2>/dev/null; then echo "Error: daemon exited during startup (bad model path?)" >&2 exit 1 fi sleep 1 waited=$((waited + 1)) if [ "$waited" -ge 120 ]; then echo "Error: timed out waiting for daemon socket at $SOCK_PATH" >&2 exit 1 fi done echo "Daemon ready (pid $DAEMON_PID) after ${waited}s" echo "Evaluating $total_prompts prompt(s) from '$PROMPT_FILE'" echo "Writing results to: '$RESULT_FILE'" echo "---------------------------------------------------" current=0 first=1 # Temporary file to capture time / rusage output TIME_LOG=$(mktemp /tmp/lex_time.XXXXXX) # Process each prompt while IFS= read -r prompt || [ -n "$prompt" ]; do case "$prompt" in ""|\#*) continue ;; esac current=$((current + 1)) if [ "$first" -eq 1 ]; then first=0 else printf "%s\n\n" "$RULE" >> "$RESULT_FILE" fi printf "PROMPT: %s\n\n" "$prompt" >> "$RESULT_FILE" # Sample daemon RSS prior to request daemon_rss_kb=$(ps -o rss= -p "$DAEMON_PID" 2>/dev/null | tr -d ' ') # Capture baseline swap occupancy before sending the prompt swap_start_mb=$(get_swap_occupancy_mb) # Send prompt to daemon over socket /usr/bin/time -lp sh -c "printf '%s\n' \"\$1\" | nc -U \"\$2\"" _ "$prompt" "$SOCK_PATH" \ >> "$RESULT_FILE" 2> "$TIME_LOG" # Capture ending swap occupancy after prompt completes swap_end_mb=$(get_swap_occupancy_mb) real_time=$(awk '/^real/ {print $2}' "$TIME_LOG") # Calculate net change in swap occupancy during request execution if [ -n "$swap_start_mb" ] && [ -n "$swap_end_mb" ]; then swap_delta_mb=$(awk -v start="$swap_start_mb" -v end="$swap_end_mb" \ 'BEGIN { printf "%+.2f", end - start }') else swap_end_mb="unavailable" swap_delta_mb="+0.00" fi if [ -n "$daemon_rss_kb" ]; then daemon_rss_mb=$(awk -v kb="$daemon_rss_kb" 'BEGIN{printf "%.2f", kb/1024}') metrics_summary="ELAPSED: ${real_time}s | DAEMON RSS: ${daemon_rss_mb} MB | SWAP: ${swap_end_mb} MB (${swap_delta_mb} MB)" else metrics_summary="ELAPSED: ${real_time}s | DAEMON RSS: unavailable | SWAP OCCUPIED: ${swap_end_mb} MB (${swap_delta_mb} MB)" fi printf "METRICS: %s\n\n" "$metrics_summary" >> "$RESULT_FILE" printf "[%d/%d] Done: %s (%s)\n" "$current" "$total_prompts" "$prompt" "$metrics_summary" done < "$PROMPT_FILE" echo "---------------------------------------------------" echo "Done! Processed $total_prompts prompt(s)."