check_global_endpoint.sh

#!/bin/bash
#
# Check that current Google models on Vertex AI work with the global endpoint.
#
# Usage:
#     ./check_global_endpoint.sh PROJECT_ID
#
# Prerequisites:
#     - gcloud CLI installed and authenticated
#     - curl, jq
#
# Reference:
#     https://cloud.google.com/vertex-ai/generative-ai/docs/learn/model-versions

set -euo pipefail

PROJECT_ID="${1:-}"
if [[ -z "$PROJECT_ID" ]]; then
    echo "Usage: $0 PROJECT_ID"
    exit 1
fi

ENDPOINT="https://aiplatform.googleapis.com/v1/projects/${PROJECT_ID}/locations/global/publishers/google/models"
TOKEN=$(gcloud auth print-access-token)

# Current stable Google models on Vertex AI (as of 2025-12)
# Reference: https://cloud.google.com/vertex-ai/generative-ai/docs/learn/model-versions
#
# Note: gemini-2.5-pro, gemini-2.5-flash, and gemini-3-pro-preview use "thinking" mode
# which consumes tokens for internal reasoning. We need sufficient maxOutputTokens to
# get a visible response.
LLM_MODELS=(
    "gemini-2.5-pro"
    "gemini-2.5-flash"
    "gemini-2.5-flash-lite"
    "gemini-2.0-flash-001"
    "gemini-2.0-flash-lite-001"
    # Preview
    "gemini-3-pro-preview"
)

EMBEDDING_MODELS=(
    "gemini-embedding-001"
    "text-embedding-005"
    "text-multilingual-embedding-002"
)

echo "Project: ${PROJECT_ID}"
echo "Location: global"
echo "Endpoint: aiplatform.googleapis.com"
echo ""

FAILED=0

check_llm() {
    local model="$1"
    local url="${ENDPOINT}/${model}:generateContent"
    local response
    response=$(curl -s -X POST "$url" \
        -H "Authorization: Bearer ${TOKEN}" \
        -H "Content-Type: application/json" \
        -d '{
            "contents": [{"role": "user", "parts": [{"text": "Say ok"}]}],
            "generationConfig": {"maxOutputTokens": 200, "temperature": 0}
        }' 2>&1)

    if echo "$response" | jq -e '.candidates[0].content.parts[0].text' > /dev/null 2>&1; then
        local text
        text=$(echo "$response" | jq -r '.candidates[0].content.parts[0].text' | head -c 30)
        echo "  ✓ ${model}: ${text}"
    else
        local error
        error=$(echo "$response" | jq -r '.error.message // .error // "unknown error"' 2>/dev/null | head -c 60)
        echo "  ✗ ${model}: ${error}"
        FAILED=1
    fi
}

check_embedding() {
    local model="$1"
    local url="${ENDPOINT}/${model}:predict"
    local response
    response=$(curl -s -X POST "$url" \
        -H "Authorization: Bearer ${TOKEN}" \
        -H "Content-Type: application/json" \
        -d '{
            "instances": [{"content": "Hello, world!"}]
        }' 2>&1)

    if echo "$response" | jq -e '.predictions[0].embeddings.values' > /dev/null 2>&1; then
        local dim
        dim=$(echo "$response" | jq '.predictions[0].embeddings.values | length')
        echo "  ✓ ${model}: dim=${dim}"
    else
        local error
        error=$(echo "$response" | jq -r '.error.message // .error // "unknown error"' 2>/dev/null | head -c 60)
        echo "  ✗ ${model}: ${error}"
        FAILED=1
    fi
}

echo "=== LLM Models ==="
for model in "${LLM_MODELS[@]}"; do
    check_llm "$model"
done

echo ""
echo "=== Embedding Models ==="
for model in "${EMBEDDING_MODELS[@]}"; do
    check_embedding "$model"
done

echo ""
if [[ $FAILED -eq 0 ]]; then
    echo "All models passed!"
    exit 0
else
    echo "Some models failed."
    exit 1
fi