#!/usr/bin/env bash
# Validation test for llama.cpp DFlash+MTP server
# Sends a simple prompt and measures generation speed (tokens/sec)
set -euo pipefail

PORT="${LLAMA_PORT:-8080}"
HOST="${LLAMA_HOST:-127.0.0.1}"
MAX_TOKENS="${MAX_TOKENS:-1024}"

echo "=== llama.cpp Validation Test ==="
echo "Target: http://${HOST}:${PORT}"
echo "Max tokens: ${MAX_TOKENS}"
echo ""

# Wait for server to be ready
echo "Waiting for server..."
for i in $(seq 1 300); do
    if curl -sf "http://${HOST}:${PORT}/health" > /dev/null 2>&1; then
        echo "Server ready after $((i * 2))s"
        break
    fi
    if [ "$i" -eq 300 ]; then
        echo "FAIL: Server did not become ready in 600s"
        exit 1
    fi
    sleep 2
done

# Send two parallel requests and capture timing
echo ""
echo "Sending 2 parallel validation prompts..."
START_TIME=$(date +%s%N)

RESPONSE1=$(curl -s \
    -X POST "http://${HOST}:${PORT}/v1/chat/completions" \
    -H "Content-Type: application/json" \
    -d "{
        \"messages\": [
            {\"role\": \"user\", \"content\": \"Explain the difference between supervised, unsupervised, and reinforcement learning in machine learning. For each type, provide a concrete real-world example, describe the training process, and discuss the advantages and limitations. Then compare them across three dimensions: data requirements, interpretability of results, and computational cost. Finally, describe a hybrid scenario where combining two of these approaches would outperform using either alone.\"}
        ],
        \"max_tokens\": ${MAX_TOKENS},
        \"temperature\": 0.7,
        \"top_p\": 0.95
    }" &)

RESPONSE2=$(curl -s \
    -X POST "http://${HOST}:${PORT}/v1/chat/completions" \
    -H "Content-Type: application/json" \
    -d "{
        \"messages\": [
            {\"role\": \"user\", \"content\": \"Write a detailed analysis of the geopolitical consequences of the fall of the Byzantine Empire in 1453. Trace the chain of events from the Ottoman conquest of Constantinople through the next two centuries, covering at minimum: the shift in Mediterranean trade routes, the Italian Renaissance's relationship with fleeing Greek scholars, the Russian claim to Roman succession, and the papal response. Then compare how the region might have developed differently had Constantinople remained independent.\"}
        ],
        \"max_tokens\": ${MAX_TOKENS},
        \"temperature\": 0.7,
        \"top_p\": 0.95
    }" &)

wait

END_TIME=$(date +%s%N)

# Extract token counts and timing from responses
COMPLETION_TOKENS1=$(echo "$RESPONSE1" | python3 -c "import sys,json; d=json.load(sys.stdin); print(d['usage']['completion_tokens'])" 2>/dev/null || echo "0")
COMPLETION_TOKENS2=$(echo "$RESPONSE2" | python3 -c "import sys,json; d=json.load(sys.stdin); print(d['usage']['completion_tokens'])" 2>/dev/null || echo "0")
TOTAL_TOKENS=$((COMPLETION_TOKENS1 + COMPLETION_TOKENS2))
TOTAL_TIME_MS=$(( (END_TIME - START_TIME) / 1000000 ))

if [ "$TOTAL_TOKENS" -eq 0 ]; then
    echo "FAIL: No tokens generated or response parse error"
    echo "Response1: $RESPONSE1"
    echo "Response2: $RESPONSE2"
    exit 1
fi

# Calculate tokens/sec
TOK_PER_SEC=$(python3 -c "print(f'{${TOTAL_TOKENS} * 1000 / ${TOTAL_TIME_MS}:.2f}')")

echo ""
echo "=== Results ==="
echo "Prompt 1 tokens: ${COMPLETION_TOKENS1}"
echo "Prompt 2 tokens: ${COMPLETION_TOKENS2}"
echo "Total tokens: ${TOTAL_TOKENS}"
echo "Total time: ${TOTAL_TIME_MS}ms"
echo "Speed: ${TOK_PER_SEC} tok/s"
echo ""

# Check minimum speed threshold
MIN_SPEED="${MIN_SPEED:-5.0}"
PASS=$(python3 -c "print('yes' if ${TOK_PER_SEC} >= ${MIN_SPEED} else 'no')")

if [ "$PASS" = "yes" ]; then
    echo "PASS: Speed ${TOK_PER_SEC} tok/s >= ${MIN_SPEED} tok/s threshold"
    exit 0
else
    echo "FAIL: Speed ${TOK_PER_SEC} tok/s < ${MIN_SPEED} tok/s threshold"
    exit 1
fi