Cette page présente la méthode de réponse en streaming.
La méthode de réponse en flux continu présente de nombreuses fonctionnalités identiques à la méthode answer, ainsi qu'une fonctionnalité supplémentaire : streaming. Lorsque vous diffusez une réponse, celle-ci est divisée en plusieurs parties qui sont envoyées séquentiellement.
L'affichage progressif des réponses est particulièrement utile si les réponses générées sont longues, car l'envoi de la réponse entière en une seule fois entraîne un délai. La diffusion en streaming des réponses réduit l'impression de latence.
Limites
La méthode de réponse en flux continu présente les mêmes caractéristiques que la méthode de réponse, à l'exception des suivantes :
Le nombre d'étapes de reformulation est de un. Vous ne pouvez pas désactiver la reformulation ni modifier le nombre maximal d'étapes.
Seuls les modèles Gemini peuvent être utilisés avec la méthode de réponse en streaming. Pour obtenir la liste des modèles, consultez Modèles disponibles.
Diffuser une réponse
La commande suivante montre comment appeler la méthode streaming answer et renvoyer une réponse générée sous la forme d'une série de réponses JSON. En règle générale, chaque réponse contient une phrase.
Cette commande de base n'affiche que les entrées requises. Les options sont laissées à leurs valeurs par défaut.
Pour obtenir des exemples d'autres options, consultez Obtenir des réponses et des informations complémentaires. Certaines options de réponse ne sont pas disponibles pour le streaming des réponses. Consultez les limitations sur cette page.
REST
Pour effectuer une recherche et obtenir des résultats avec une réponse générée en streaming :
Exécutez la commande curl suivante :
curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \ -H "Content-Type: application/json" \ "https://discoveryengine.googleapis.com/v1/projects/PROJECT_ID/locations/global/collections/default_collection/engines/APP_ID/servingConfigs/default_search:streamAnswer" \ -d '{ "query": { "text": "QUERY"} }'Remplacez les éléments suivants :
PROJECT_ID: ID de votre projet Google Cloud .APP_ID: ID de l'application Vertex AI Search que vous souhaitez interroger.QUERY: chaîne de texte libre contenant la question ou la requête de recherche. Par exemple, "Quelle base de données est la plus rapide, BigQuery ou Spanner ?".
Exemple de commande et de résultat partiel
curl -X POST -H "Authorization: Bearer $(gcloud auth print-access-token)" \ -H "Content-Type: application/json" \ "https://discoveryengine.googleapis.com/v1/projects/my-project-123/locations/global/collections/default_collection/engines/my-app/servingConfigs/default_search:streamAnswer" \ -d '{ "query":{"text":"Which database is faster, bigquery or spanner?"} }'
[{ "answer": { "state": "STREAMING", "steps": [ { "description": "Rephrase the query and search.", "actions": [ { "searchAction": { "query": " What is the performance of Spanner?" }, "observation": { "searchResults": [ { "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/1a9f55e00c42c06ca97bf5a5868dbcdc", "uri": "https://cloud.google.com/generative-ai-app-builder/docs/answer", "title": "Get answers and follow-ups", "snippetInfo": [ { "snippet": "QUERY : a free-text string that contains the question or search query. For example, "Compare the BigQuery and \u003cb\u003eSpanner\u003c/b\u003e databases?". Example command and result.", "snippetStatus": "SUCCESS" } ] }, ... { "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/b95bb201a0adb24f769627f56cf34405", "uri": "https://abc.xyz/assets/investor/static/pdf/2017_Q1_Earnings_Transcript.pdf", "title": "\u200b \u200b", "snippetInfo": [ { "snippet": "well as Hardware related costs, reflecting the continued strong \u003cb\u003eperformance\u003c/b\u003e of our new Made by ... We introduced dozens of new products, including \u003cb\u003eSpanner\u003c/b\u003e, a ...", "snippetStatus": "SUCCESS" } ] } ] } }, { "searchAction": { "query": " What is the performance of BigQuery?" }, "observation": { "searchResults": [ { "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/18bcc727bfd6a3d1be0aa4bd49fe2c50", "uri": "https://cloud.google.com/generative-ai-app-builder/docs/evaluate-search-quality", "title": "Evaluate search quality", "snippetInfo": [ { "snippet": "You can evaluate the \u003cb\u003eperformance\u003c/b\u003e of generic search apps that contain structured, unstructured, and website data. ... Import from \u003cb\u003eBigQuery\u003c/b\u003e: import \u003cb\u003eBigQuery\u003c/b\u003e data ...", "snippetStatus": "SUCCESS" } ] }, ... { "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/1a9f55e00c42c06ca97bf5a5868dbcdc", "uri": "https://cloud.google.com/generative-ai-app-builder/docs/answer", "title": "Get answers and follow-ups", "snippetInfo": [ { "snippet": "QUERY : a free-text string that contains the question or search query. For example, "Compare the \u003cb\u003eBigQuery\u003c/b\u003e and Spanner databases?". Example command and result.", "snippetStatus": "SUCCESS" } ] } ] } } ] } ] } } , { "answer": { "state": "STREAMING", "references": [ { "chunkInfo": { "content": "Example command and partial result curl -X POST -H \"Authorization: Bearer $(gcloud auth print-access-token)\" -H \"Content-Type: application/json\" \"https://discoveryengine.googleapis.com/v1/projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/servingConfigs/default_search:answer\" -d '{ \"query\": { \"text\": \"Compare bigquery with spanner database?\"} \"queryUnderstandingSpec\": { \"queryRephraserSpec\": { \"disable\": true } } }' { \"answer\": { \"state\": \"SUCCEEDED\", \"answerText\": \"You can compare BigQuery and Spanner databases using the following criteria:\\n\\n* **Pricing:** BigQuery is priced per GB of data processed, while Spanner is priced per hour of compute time.\\n* **Performance:** BigQuery is designed for fast analytics, while Spanner is designed for high availability and scalability.\\n* **Features:** BigQuery supports a wide range of features, including SQL, machine learning, and streaming. ", "documentMetadata": { "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/1a9f55e00c42c06ca97bf5a5868dbcdc", "uri": "https://cloud.google.com/generative-ai-app-builder/docs/answer", "title": "Get answers and follow-ups" } } }, ... { "chunkInfo": { "content": "Here is an example of a summary, with citations and citation metadata, returned at the end of a search response: See more code actions. Dismiss View Light code theme Dark code theme \"summary\": { \"summaryText\": \"BigQuery is Google Cloud's fully managed and completely serverless enterprise data warehouse [1]. BigQuery supports all data types, works across clouds, and has built-in machine learning and business intelligence, all within a unified platform [2, 3].\", \"summaryWithMetadata\": { \"summary\": \"BigQuery is Google Cloud's fully managed and completely serverless enterprise data warehouse. ", "documentMetadata": { "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/f7ba2e8666f5514b5bc14f5e300d7678", "uri": "https://cloud.google.com/generative-ai-app-builder/docs/get-search-summaries", "title": "Get search summaries" } } } ] } } , { "answer": { "state": "STREAMING", "answerText": "Span" } } , { "answer": { "state": "STREAMING", "answerText": "ner is Google's large-scale database that scales 20 times better than" } } , ... { "answer": { "state": "STREAMING", "answerText": " Web Services, and on-premises data sources. " } } , { "answer": { "state": "STREAMING", "answerText": "Spanner is a distributed, strongly consistent, SQL database designed to scale to 10 million servers. \n" } } , { "answer": { "state": "SUCCEEDED", "answerText": "Spanner is Google's large-scale database that scales 20 times better than any competitor. Spanner is designed for high availability and scalability, while BigQuery is designed for fast analytics. BigQuery is a serverless, highly scalable, and cost-effective cloud data warehouse that enables businesses to analyze all their data very quickly. BigQuery is a very powerful tool that can be used to analyze data from many different sources, including Google Cloud Platform, Amazon Web Services, and on-premises data sources. Spanner is a distributed, strongly consistent, SQL database designed to scale to 10 million servers. \n", "references": [ { "chunkInfo": { "content": "Example command and partial result curl -X POST -H \"Authorization: Bearer $(gcloud auth print-access-token)\" -H \"Content-Type: application/json\" \"https://discoveryengine.googleapis.com/v1/projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/servingConfigs/default_search:answer\" -d '{ \"query\": { \"text\": \"Compare bigquery with spanner database?\"} \"queryUnderstandingSpec\": { \"queryRephraserSpec\": { \"disable\": true } } }' { \"answer\": { \"state\": \"SUCCEEDED\", \"answerText\": \"You can compare BigQuery and Spanner databases using the following criteria:\\n\\n* **Pricing:** BigQuery is priced per GB of data processed, while Spanner is priced per hour of compute time.\\n* **Performance:** BigQuery is designed for fast analytics, while Spanner is designed for high availability and scalability.\\n* **Features:** BigQuery supports a wide range of features, including SQL, machine learning, and streaming. ", "documentMetadata": { "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/1a9f55e00c42c06ca97bf5a5868dbcdc", "uri": "https://cloud.google.com/generative-ai-app-builder/docs/answer", "title": "Get answers and follow-ups" } } }, { "chunkInfo": { "content": "Second, we also give them the ability to build applications using a set of technology that can run on any environment that they have. When we say on any environment - at their premise, on our cloud or on any other cloud. So, in other words, they can learn once, write once, deploy anywhere; and we make money no matter where they deploy. An example of that is a recent product we introduced called AlloyDB. It's the fastest-performing relational database in the market. We run it in all four environments: Our cloud, on-premise and on other clouds. And it's the only relational database that can run in any of those configurations. You see that in our adoption, both at the top end of the market where a system - for example, like Spanner, which is our large-scale database - scales 20 times better than the largest scalable system of any competitor. So for high-end, we work extremely well. And, also, because we made it so easy to use, startups and small businesses are growing very quickly in their adoption of our platform. When we introduced our AI systems, we introduced a platform called Vertex AI. ", "documentMetadata": { "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/8ad08d0844d601733e135381512e2a16", "uri": "http://abc.xyz/thomas-kurian-ceo-google-cloud-at-the-goldman-sachs-2023-communacopia-technology-conference-on-september-7th-2023", "title": "Thomas Kurian, CEO, Google Cloud at the Goldman Sachs 2023 Communacopia + Technology Conference on September 7th, 2023 - Alphabet Investor Relations" } } }, ... { "chunkInfo": { "content": "BigQuery is also integrated with other Google Cloud services, such as Google Kubernetes Engine, Cloud Data Fusion, and Cloud Dataproc, making it easy to build and deploy data pipelines. Here are some of the benefits of using BigQuery: * **Fast and scalable:** BigQuery can process petabytes of data very quickly, and it can scale to handle even the most demanding workloads. * **Cost-effective:** BigQuery is a very cost-effective way to store and analyze data. You only pay for the data that you use, and there are no upfront costs or commitments. * **Secure:** BigQuery is a secure platform that meets the needs of even the most security-conscious organizations. * **Easy to use:** BigQuery is easy to use, even for non-technical users. It has a simple and intuitive user interface, and it supports a variety of data sources. * **Integrated with other Google Cloud services:** BigQuery is integrated with other Google Cloud services, making it easy to build and deploy data pipelines. If you are looking for a fast, scalable, and cost-effective way to analyze your data, then BigQuery is a great option. ", "documentMetadata": { "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/f7ba2e8666f5514b5bc14f5e300d7678", "uri": "https://cloud.google.com/generative-ai-app-builder/docs/get-search-summaries", "title": "Get search summaries" } } }, { "chunkInfo": { "content": "Here is an example of a summary, with citations and citation metadata, returned at the end of a search response: See more code actions. Dismiss View Light code theme Dark code theme \"summary\": { \"summaryText\": \"BigQuery is Google Cloud's fully managed and completely serverless enterprise data warehouse [1]. BigQuery supports all data types, works across clouds, and has built-in machine learning and business intelligence, all within a unified platform [2, 3].\", \"summaryWithMetadata\": { \"summary\": \"BigQuery is Google Cloud's fully managed and completely serverless enterprise data warehouse. ", "documentMetadata": { "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/f7ba2e8666f5514b5bc14f5e300d7678", "uri": "https://cloud.google.com/generative-ai-app-builder/docs/get-search-summaries", "title": "Get search summaries" } } } ], "steps": [ { "description": "Rephrase the query and search.", "actions": [ { "searchAction": { "query": " What is the performance of Spanner?" }, "observation": { "searchResults": [ { "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/1a9f55e00c42c06ca97bf5a5868dbcdc", "uri": "https://cloud.google.com/generative-ai-app-builder/docs/answer", "title": "Get answers and follow-ups", "snippetInfo": [ { "snippet": "QUERY : a free-text string that contains the question or search query. For example, "Compare the BigQuery and \u003cb\u003eSpanner\u003c/b\u003e databases?". Example command and result.", "snippetStatus": "SUCCESS" } ] }, { "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/9d022f7bdf24bac6714a9cf61a5458c7", "uri": "https://abc.xyz/assets/87/4c/162ca71d4178a3f4d39002467439/thomas-kurian-goldman-sachs-090723.pdf", "title": "Thomas Kurian Goldman Sachs 090723", "snippetInfo": [ { "snippet": "2X better training \u003cb\u003eperformance\u003c/b\u003e per dollar1 compared to a leading cloud alternative. More than 70% of gen AI unicorns are Google Cloud customers. Best ...", "snippetStatus": "SUCCESS" } ] }, ... { "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/20641e370fa86c78f1c81f3dab22efe1", "uri": "https://cloud.google.com/generative-ai-app-builder/docs/release-notes", "title": "AI Applications release notes | Google Cloud", "snippetInfo": [ { "snippet": "Generative answers have been updated with \u003cb\u003eperformance\u003c/b\u003e improvements. ... This lets you assess your search engine's \u003cb\u003eperformance\u003c/b\u003e ... Importing data from \u003cb\u003eSpanner\u003c/b\u003e, Cloud ...", "snippetStatus": "SUCCESS" } ] }, { "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/b95bb201a0adb24f769627f56cf34405", "uri": "https://abc.xyz/assets/investor/static/pdf/2017_Q1_Earnings_Transcript.pdf", "title": "\u200b \u200b", "snippetInfo": [ { "snippet": "well as Hardware related costs, reflecting the continued strong \u003cb\u003eperformance\u003c/b\u003e of our new Made by ... We introduced dozens of new products, including \u003cb\u003eSpanner\u003c/b\u003e, a ...", "snippetStatus": "SUCCESS" } ] } ] } }, { "searchAction": { "query": " What is the performance of BigQuery?" }, "observation": { "searchResults": [ { "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/18bcc727bfd6a3d1be0aa4bd49fe2c50", "uri": "https://cloud.google.com/generative-ai-app-builder/docs/evaluate-search-quality", "title": "Evaluate search quality", "snippetInfo": [ { "snippet": "You can evaluate the \u003cb\u003eperformance\u003c/b\u003e of generic search apps that contain structured, unstructured, and website data. ... Import from \u003cb\u003eBigQuery\u003c/b\u003e: import \u003cb\u003eBigQuery\u003c/b\u003e data ...", "snippetStatus": "SUCCESS" } ] }, { "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/2a3221d40533a4bdaf35778962a2a079", "uri": "https://cloud.google.com/generative-ai-app-builder/docs/check-media-data-quality", "title": "Check data quality for media recommendations", "snippetInfo": [ { "snippet": "... model that will result in \u003cb\u003eperformance\u003c/b\u003e issue if not met for all media recommendations models and all business objectives.", "condition": { "expression ...", "snippetStatus": "SUCCESS" } ] }, ... { "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/18c258b9c770f4d762e6233d1a1bc81c", "uri": "https://cloud.google.com/generative-ai-app-builder/docs/user-events", "title": "About user events", "snippetInfo": [ { "snippet": "This section provides the data formats for each event type supported by media recommendations. Examples for JavaScript Pixel are provided. For \u003cb\u003eBigQuery\u003c/b\u003e, the ...", "snippetStatus": "SUCCESS" } ] }, { "document": "projects/123456/locations/global/collections/default_collection/dataStores/my-data-store/branches/0/documents/1a9f55e00c42c06ca97bf5a5868dbcdc", "uri": "https://cloud.google.com/generative-ai-app-builder/docs/answer", "title": "Get answers and follow-ups", "snippetInfo": [ { "snippet": "QUERY : a free-text string that contains the question or search query. For example, "Compare the \u003cb\u003eBigQuery\u003c/b\u003e and Spanner databases?". Example command and result.", "snippetStatus": "SUCCESS" } ] } ] } } ] } ] } }Dans cet exemple, la réponse à la requête "Quelle base de données est la plus rapide, BigQuery ou Spanner ?" s'affiche dans une série de sorties JSON. La sortie finale est associée à l'état
SUCCEEDEDet inclut la réponse complète.Dans cet exemple, les réponses de streaming
stepsetreferencesapparaissent avant les réponses de streamingAnswerText. Ce n'est pas toujours le cas. Si vous analysez la sortie, ne partez pas du principe que les réponsesstepsetreferencesprécèdent les réponsesAnswerText.
Autres exemples
La commande de base présentée dans Diffuser une réponse en streaming est la commande la plus simple, sans aucune option spécifiée. Toutefois, vous pouvez appliquer les mêmes options que celles disponibles avec la méthode answer, à l'exception des limitations listées sur cette page.
Les réponses en streaming peuvent également être utilisées avec les sessions de suivi.