#!/bin/sh
# Capture a non-streaming local OpenAI-compatible completion measurement.
set -eu

if [ "$#" -ne 4 ]; then
    echo "usage: $0 BASE_URL MODEL PROMPT_FILE OUTPUT_JSON" >&2
    exit 1
fi

base_url=$1
model=$2
prompt_file=$3
output=$4

if [ ! -r "$prompt_file" ]; then
    echo "gem: prompt file is not readable: $prompt_file" >&2
    exit 1
fi

request=$(jq -n --rawfile prompt "$prompt_file" --arg model "$model" \
    '{model: $model, messages: [{role: "user", content: $prompt}], max_tokens: 128, stream: false}')
metrics=$(mktemp /private/tmp/gem-server-probe.XXXXXX)
trap 'rm -f "$metrics"' EXIT

curl --fail --silent --show-error --output "$output" --write-out '%{time_starttransfer} %{time_total}\n' \
    -H 'Content-Type: application/json' --data "$request" "$base_url/chat/completions" > "$metrics"

read -r ttfb total < "$metrics"
printf 'ttfb_seconds=%s\ntotal_seconds=%s\nresponse=%s\n' "$ttfb" "$total" "$output"