跳转至

文档中心

MaaS_Ge batch processing

MaaS_Ge batch processing

google文档地址：https://docs.cloud.google.com/vertex-ai/generative-ai/docs/multimodal/batch-prediction-from-cloud-storage?hl=zh-cn

一、上传文件

说明：文件最大支持100条请求数据，且文件最大不超过50MB，文件固定保存7天

批处理文件（jsonl格式）

batch_requests_for_multimodal_input_2.jsonl

文件内容示例

{"request":{"contents": [{"role": "user", "parts": [{"text": "List objects in this image."}, {"file_data": {"file_uri": "gs://cloud-samples-data/generative-ai/image/office-desk.jpeg", "mime_type": "image/jpeg"}}]}],"generationConfig":{"temperature": 0.4}}}
{"request":{"contents": [{"role": "user", "parts": [{"text": "List objects in this image."}, {"file_data": {"file_uri": "gs://cloud-samples-data/generative-ai/image/gardening-tools.jpeg", "mime_type": "image/jpeg"}}]}],"generationConfig":{"temperature": 0.4}}}

入参：

参数名	类型	是否必填	描述
name	string	是	文件名
uploadType	string	是	固定值media

curl

```plain text curl --location --request POST 'https://genaiapi.cloudsway.net/v1/ai/batch/gemini/{ENDPOINT}/files?uploadType=media&name=ImageToImage.jsonl' \ --header 'Authorization: Bearer {KEY}' \ --header 'Content-Type: application/octet-stream' \ --data-binary '@C:\Users\Desktop\batch\gemini\ImageToImage.jsonl'

### 出参

```json
{
    "kind": "storage#object",
    "id": "1779785513731222",
    "name": "ImageToImage.jsonl",
    "generation": "1779785513731222",
    "metageneration": "1",
    "contentType": "application/octet-stream",
    "storageClass": "STANDARD",
    "size": "510",
    "md5Hash": "pM5qF/A+Eetto7iq5PZMsA==",
    "crc32c": "unFIEg==",
    "etag": "CJaJytrJ1pQDEAE=",
    "timeCreated": "2026-05-26T08:51:53.742Z",
    "updated": "2026-05-26T08:51:53.742Z",
    "timeStorageClassUpdated": "2026-05-26T08:51:53.742Z",
    "timeFinalized": "2026-05-26T08:51:53.742Z"
}

二、创建批处理任务

入参：

参数名	层级	类型	是否必填	描述
displayName	1	string	是	任务名称
inputConfig	1	object	是
instancesFormat	2	string	是	固定值jsonl
gcsSource	2	object	是
uris	3	string	是	上传文件的name值（第一步出参的name值）
outputConfig	1	object	是
predictionsFormat	2	string	是	固定值jsonl

curl

curl --location --request POST 'https://genaiapi.cloudsway.net/v1/ai/batch/gemini/{ENDPOINT}/batchPredictionJobs' \
--header 'Authorization: Bearer {KEY}' \
--header 'Content-Type: application/json' \
--data-raw '{
  "displayName": "my-cloud-storage-batch-inference-job",
  "inputConfig": {
    "instancesFormat": "jsonl",
    "gcsSource": {
      "uris": "{name}"
    }
  },
  "outputConfig": {
    "predictionsFormat": "jsonl"
  }
}'

出参

{
    "name": "903523130369114112",
    "displayName": "my-cloud-storage-batch-inference-job",
    "model": "MaaS_Ge_3.1_flash_image_preview_20260226",
    "inputConfig": {
        "instancesFormat": "jsonl",
        "gcsSource": {
            "uris": [
                "ImageToImage.jsonl"
            ]
        }
    },
    "outputConfig": {
        "predictionsFormat": "jsonl",
        "gcsDestination": {
            "outputUriPrefix": "response"
        }
    },
    "state": "JOB_STATE_PENDING",
    "createTime": "2026-05-28T08:39:00.803441Z",
    "updateTime": "2026-05-28T08:39:00.803441Z",
    "modelVersionId": "1"
}

三、批处理状态

入参：

参数名	类型	是否必填	描述
name	string	是	第二步出参的name值

curl

curl --location --request GET 'https://genaiapi.cloudsway.net/v1/ai/batch/gemini/{ENDPOINT}/batchPredictionJobs/{name}' \
--header 'Authorization: Bearer {KEY}'

出参

{
    "name": "4787616125208756224",
    "displayName": "my-cloud-storage-batch-inference-job",
    "model": "MaaS_Ge_3_pro_image_preview_20251118_batch",
    "inputConfig": {
        "instancesFormat": "jsonl",
        "gcsSource": {
            "uris": [
                "MaaS_Gemini_batch_1780628300511.jsonl"
            ]
        }
    },
    "outputConfig": {
        "predictionsFormat": "jsonl",
        "gcsDestination": {
            "outputUriPrefix": "response"
        }
    },
    "outputInfo": {
        "gcsOutputDirectory": "https://genaiapi.cloudsway.net/v1/ai/batch/gemini/{ENDPOINT}/batchPredictionJobs/4787616125208756224/content"
    },
    "state": "JOB_STATE_SUCCEEDED",
    "error": null,
    "completionStats": {
        "successfulCount": "3",
        "failedCount": null
    },
    "createTime": "2026-06-05T02:58:27.175092Z",
    "startTime": "2026-06-05T02:59:01.122705Z",
    "endTime": "2026-06-05T03:01:26.124246Z",
    "updateTime": "2026-06-05T03:01:26.124246Z",
    "modelVersionId": "1"
}

四、获取批处理结果

说明：返回结果固定保存7天

入参：

参数名	类型	是否必填	描述
alt	string	否	当传json时，返回结果的元数据信息，当传media时，返回的是执行结果数据。默认为json
name	string	是	第二步出参的name值

curl

curl --location --request GET 'https://genaiapi.cloudsway.net/v1/ai/batch/gemini/{ENDPOINT}/batchPredictionJobs/{name}/content?alt=media' \
--header 'Authorization: Bearer {KEY}'

出参（alt为media）

{"request":{"contents":[{"role":"user","parts":[{"file_data":null,"text":"List objects in this image."},{"file_data":{"file_uri":"gs://cloud-samples-data/generative-ai/image/office-desk.jpeg","mime_type":"image/jpeg"},"text":null}]}],"generationConfig":{"temperature":0.4}},"response":{"modelVersion":"MaaS_Ge_2.5_flash_20250617_batch","candidates":[{"content":{"role":"model","parts":[{"text":"Here are the objects visible in the image:\n\n*   Globe\n*   Eiffel Tower figurine\n*   Airplane model\n*   Tablet (with a blank white screen)\n*   Miniature shopping cart\n*   Red gift box (inside the shopping cart)\n*   Coffee cup\n*   Saucer\n*   Coffee (in the cup)\n*   Keyboard\n*   Computer mouse\n*   Passport\n*   Sunglasses\n*   One-dollar bill\n*   Notebook / Notepad\n*   Pen"}]},"score":-75.72122192382813,"avgLogprobs":-0.7011224252206308,"finishReason":"STOP"}],"createTime":"2026-01-29T02:15:37.871417Z","usageMetadata":{"candidatesTokenCount":108,"totalTokenCount":2163,"trafficType":"ON_DEMAND","billablePromptUsage":{"imageCount":1,"textCount":23},"promptTokensDetails":[{"modality":"TEXT","tokenCount":6},{"modality":"IMAGE","tokenCount":1806}],"thoughtsTokenCount":243,"candidatesTokensDetails":[{"modality":"TEXT","tokenCount":108}],"promptTokenCount":1812},"responseId":"ycJ6afmXNYfWitUPtNOwuQE"},"processed_time":"2026-01-29T02:15:47.048543+00:00","status":""}
{"request":{"contents":[{"role":"user","parts":[{"file_data":null,"text":"List objects in this image."},{"file_data":{"file_uri":"gs://cloud-samples-data/generative-ai/image/gardening-tools.jpeg","mime_type":"image/jpeg"},"text":null}]}],"generationConfig":{"temperature":0.4}},"response":{"modelVersion":"MaaS_Ge_2.5_flash_20250617_batch","candidates":[{"content":{"role":"model","parts":[{"text":"Here are the objects visible in the image:\n\n1.  **Watering can:** A large green plastic watering can with a white sprinkler head (rose).\n2.  **Potted plant:** A small green plant growing in a terracotta pot.\n3.  **Terracotta pots:** Two empty terracotta pots, stacked one on top of the other.\n4.  **Hand trowel:** A small green gardening shovel with a black handle.\n5.  **Hand hoe/cultivator:** A small green gardening tool with a hoe on one side and a three-pronged cultivator on the other, with a black handle.\n6.  **Gardening gloves:** A pair of striped green, yellow, and white fabric gloves.\n7.  **Grass:** The green lawn serving as the background surface."}]},"score":-168.0503692626953,"avgLogprobs":-1.0002998170398532,"finishReason":"STOP"}],"createTime":"2026-01-29T02:15:30.858407Z","usageMetadata":{"candidatesTokenCount":168,"totalTokenCount":2832,"trafficType":"ON_DEMAND","billablePromptUsage":{"imageCount":1,"textCount":23},"promptTokensDetails":[{"modality":"TEXT","tokenCount":6},{"modality":"IMAGE","tokenCount":1806}],"thoughtsTokenCount":852,"candidatesTokensDetails":[{"modality":"TEXT","tokenCount":168}],"promptTokenCount":1812},"responseId":"wsJ6aaeyNIXVitUP5ruJiAo"},"processed_time":"2026-01-29T02:15:39.343388+00:00","status":""}

出参（alt为json）

{
    "kind": "storage#object",
    "id": "1780628480806264",
    "generation": "1780628480806264",
    "metageneration": "1",
    "contentType": "application/octet-stream",
    "storageClass": "STANDARD",
    "size": "13306514",
    "crc32c": "40/N2Q==",
    "componentCount": 1,
    "etag": "CPiazYCO75QDEAE=",
    "timeCreated": "2026-06-05T03:01:20.809Z",
    "updated": "2026-06-05T03:01:20.809Z",
    "timeStorageClassUpdated": "2026-06-05T03:01:20.809Z",
    "timeFinalized": "2026-06-05T03:01:20.809Z"
}