Kong GatewayのAI ProxyプラグインでAmazon Bedrockのモデルを実行してみる

{
    "Version": "2012-10-17",
    "Statement": [
        {
            "Effect": "Allow",
            "Action": [
                "bedrock:InvokeModel"
            ],
            "Resource": "*"
        }
    ]
}

# システムの更新
sudo dnf update -y

# Dockerのインストール
sudo dnf install docker -y

# Dockerサービスの起動
sudo systemctl start docker

# システム起動時にDockerを自動起動する設定
sudo systemctl enable docker

# 現在のユーザー(ec2-user)をdockerグループに追加
sudo usermod -aG docker ec2-user

# Dockerバージョンの確認
docker --version

curl -i -X POST http://{xxx.xxx.xxx.xxx}:8000/bedrock/novalite \
-H 'Content-Type: application/json' \
--data-raw '{ "messages": [ { "role": "system", "content": "あなたは数学者です" }, { "role": "user", "content": "1 + 1はなんですか？"} ] }'

HTTP/1.1 200 OK
Content-Type: application/json
Transfer-Encoding: chunked
Connection: keep-alive
Date: Mon, 09 Dec 2024 05:55:06 GMT
x-amzn-RequestId: 031e9356-d759-498f-9345-891c4861b152
X-Kong-LLM-Model: bedrock/amazon.nova-lite-v1:0
X-Kong-Upstream-Latency: 249
X-Kong-Proxy-Latency: 1
Via: 1.1 kong/3.8.0.0-enterprise-edition
Server: kong/3.8.0.0-enterprise-edition
X-Kong-Request-Id: 7287413cabbf3741d1eaa03dab06216b

{
	"object": "chat.completion",
	"model": "amazon.nova-lite-v1:0",
	"usage": {
		"prompt_tokens": 14,
		"total_tokens": 71,
		"completion_tokens": 57
	},
	"choices": [
		{
			"finish_reason": "end_turn",
			"message": {
				"content": "1 + 1は2です。これは基本的な加算の法則に基づくもので、2つの1を合計すると2になります。数学では、このような基本的な演算は非常に重要で、より複雑な計算や理論の基礎となります。",
				"role": "assistant"
			},
			"index": 0
		}
	]
}

curl -i -X POST http://{xxx.xxx.xxx.xxx}:8000/bedrock/novalite \
-H 'Content-Type: application/json' \
--data-raw '{ "messages": [ { "role": "system", "content": "あなたは数学者です" }, { "role": "user", "content": "1 + 1はなんですか？"} ] }'

HTTP/1.1 400 Bad Request
Date: Mon, 09 Dec 2024 06:39:47 GMT
Content-Type: application/json; charset=utf-8
Connection: keep-alive
Content-Length: 35
X-Kong-Response-Latency: 1
Server: kong/3.8.0.0-enterprise-edition
X-Kong-Request-Id: 8d470f3ff7bf486f03b6e443f17b4e1c

{"error":{"message":"bad request"}}

curl -i -X POST http://{xxx.xxx.xxx.xxx}:8000/bedrock/novalite \
-H 'Content-Type: application/json' \
--data-raw '{ "messages": [ { "role": "system", "content": "あなたは数学者です" }, { "role": "user", "content": "こんにちは、1 + 1はなんですか？"} ] }'

{
	"object": "chat.completion",
	"model": "amazon.nova-lite-v1:0",
	"usage": {
		"prompt_tokens": 16,
		"total_tokens": 45,
		"completion_tokens": 29
	},
	"choices": [
		{
			"finish_reason": "end_turn",
			"message": {
				"content": "こんにちは、1 + 1 は 2 です。もっと複雑な計算や数学に関する質問があれば、いつでも聞いてください。",
				"role": "assistant"
			},
			"index": 0
		}
	]
}

curl -i -X POST http://{xxx.xxx.xxx.xxx}:8000/bedrock/haiku \
-H 'Content-Type: application/json' \
--data-raw '{ "messages": [ { "role": "system", "content": "あなたは数学者です" }, { "role": "user", "content": "1 + 1はなんですか？"} ] }'

HTTP/1.1 200 OK
Content-Type: application/json
Transfer-Encoding: chunked
Connection: keep-alive
Date: Mon, 09 Dec 2024 06:06:08 GMT
x-amzn-RequestId: 2042351d-2100-4059-b6be-333d278f9aee
X-Kong-LLM-Model: bedrock/anthropic.claude-3-haiku-20240307-v1:0
X-Kong-Upstream-Latency: 432
X-Kong-Proxy-Latency: 2
Via: 1.1 kong/3.8.0.0-enterprise-edition
Server: kong/3.8.0.0-enterprise-edition
X-Kong-Request-Id: 0ef399731b1f2659c759542a7144e172

{
	"object": "chat.completion",
	"model": "anthropic.claude-3-haiku-20240307-v1:0",
	"usage": {
		"prompt_tokens": 27,
		"total_tokens": 43,
		"completion_tokens": 16
	},
	"choices": [
		{
			"finish_reason": "end_turn",
			"message": {
				"content": "1 + 1 は 2 です。",
				"role": "assistant"
			},
			"index": 0
		}
	]
}

curl -i -X POST http://{xxx.xxx.xxx.xxx}:8000/bedrock/haiku
-H 'Content-Type: application/json' \
--data-raw '{ "messages": [ { "role": "system", "content": "あなたは数学者です" }, { "role": "user", "content": "1 + 1はなんですか？"} ] }'

HTTP/1.1 400 Bad Request
Content-Type: application/json
Connection: keep-alive
Date: Mon, 09 Dec 2024 06:01:27 GMT
x-amzn-RequestId: 42436ca7-795e-4a31-b789-d2342510ffac
x-amzn-ErrorType: ValidationException:http://internal.amazon.com/coral/com.amazon.bedrock/
Content-Length: 210
X-Kong-Upstream-Latency: 30
X-Kong-Proxy-Latency: 3
Via: 1.1 kong/3.8.0.0-enterprise-edition
Server: kong/3.8.0.0-enterprise-edition
X-Kong-Request-Id: d59e1b24f6d59e04a9c7543f83ee3714

{
	"message": "Invocation of model ID anthropic.claude-3-5-haiku-20241022-v1:0 with on-demand throughput isn’t supported. Retry your request with the ID or ARN of an inference profile that contains this model."
}