Calling Atomic Services for Inference

Download

Mode fokus

Ukuran font

Terakhir diperbarui: 2026-08-25 14:37:07

Diterjemahkan oleh AI

Tencent Cloud ES provides a rich set of atomic services (see here). You can view them by clicking "Atomic Services" under "Intelligent Search Development" in the left sidebar of the ES console. They can be called directly through APIs.
Among them, atomic services such as Embedding, Rerank, and LLM can be created as inference endpoints through the Elasticsearch inference feature.
﻿
Creating an Inference Endpoint from an Atomic Service
You can create an atomic service as an Elasticsearch inference endpoint. This allows you to call it directly through inference APIs in Elasticsearch when needed, for example, when writing data through an ingest pipeline or performing searches using k-nearest neighbor (kNN) or semantic search.
PUT /_inference/text_embedding/tencentcloudapi_bge_base_zh-v1.5" 
{
    "service": "tencent_cloud_ai_search",  
    "service_settings": {
        "secret_id": "AKIDI***********************",
        "secret_key": "BQa*************************", 
        "url": "https://es.internal.tencentcloudapi.com",
        "model_id": "bge-base-zh-v1.5",
        "region":  "ap-beijing",   
        "language":  "zh-CN",
        "action":  "GetTextEmbedding",
        "version":  "2025-01-01"
    }
}
 
Remark 1: Keys use the secret_id and secret_key format. Ensure that permissions of the TencentCloud API key include the permissions for calling Elasticsearch APIs.
Remark 2: For url, it should start with http/https. By default, you do not need to modify it.
Remark 3: Fill in the model name for model_id, for example, bge-base-zh-v1.5 or Conan-embedding-v1. Support for custom models will be added later.
Remark 4: You can fill in ap-beijing for region. Currently, atomic services are deployed only in the Beijing region.
Remark 5: Other fields usually do not need to be changed.
Reference: Overview of Atomic Services
Embedding Service
Embedding is a technique that maps high-dimensional data into a low-dimensional space. It is typically used to convert unstructured data, such as text, images, or audio, into vector representations. This makes it easier to input the data into machine models for processing. Furthermore, the distance between vectors can reflect the similarity between the objects.
For the API documentation, see Obtaining Feature Vectors.
Atomic Service
Token Limit
Dimension
Language
Remarks
bge-base-zh-v1.5
512
768
Chinese
Classic BGE model.
KaLM-embedding-multilingual-mini-v1
131072
896
Multilingual
Self-developed embedding model of WeChat based on an autoregressive LLM with high-quality training data, suitable for scenarios involving ultra-long texts, Chinese-English mixed queries, and multilingual document matching.
bge-m3
8194
1024
Multilingual
Classic BGE model.
conan-embedding-v1
512
1792
Chinese
Self-developed by Tencent, this model once ranked first overall on the MTEB leaderboard for Chinese scenarios last year.
Rerank Service
Reranking refers to the process of evaluating the relevance between documents and queries during Retrieval-Augmented Generation (RAG), placing the most relevant documents at the top of the ranking list. This ensures that the LLM prioritizes the top-ranked context when generating responses, improving the accuracy and credibility of the generated results. This method can also be used for filtering to reduce large model costs.
For the API documentation, see Reranking.
Atomic Service
Token Limit
Language
Remarks
bge-reranker-large
514
Chinese, English
Classic BGE model.
bge-reranker-v2-m3
8194
Multilingual
Classic BGE model.
Reference: Inference API Call Guide
Embedding API
Create an atomic service Embedding model in Elasticsearch, for example, named tencentcloudapi_bge_base_zh-v1.5.
(for which keys can be fully overwritten during updates)
﻿
curl  -H "Content-Type: application/json" -XPUT "http://127.0.0.1:9200/_inference/text_embedding/tencentcloudapi_bge_base_zh-v1.5" -d '{
    "service": "tencent_cloud_ai_search",  
    "service_settings": {
        "secret_id": "AKIDI***********************",
        "secret_key": "BQa*************************", 
        "url": "https://es.internal.tencentcloudapi.com",
        "model_id": "bge-base-zh-v1.5",
        "region":  "ap-beijing",   
        "language":  "zh-CN",
        "action":  "GetTextEmbedding",
        "version":  "2025-01-01"
    }
}'
Note 1: Use secret_id and secret_key for the key. Ensure that the TencentCloud API key permission includes ES API calls.
Note 2: The url must start with http/https. No modification is needed by default.
Note 3: For model_id, enter the model name, such as bge-base-zh-v1.5 or Conan-embedding-v1. Custom models will be supported later.
Note 4: For region, you can enter ap-beijing. Currently, the atomic service is deployed only in the Beijing region.
Remark 5: Other fields usually do not need to be changed.
Call the atomic service model
to perform vectorization separately
 
curl  -H "Content-Type: application/json" -XPOST "http://127.0.0.1:9200/_inference/text_embedding/tencentcloudapi_bge_base_zh-v1.5" -d '{
    "input": ["China", "United States", "United Kingdom"]
}'
View models
Obtain all models: GET /_inference/_all.
Or obtain a single model:
curl -XGET "http://127.0.0.1:9200/_inference/text_embedding/tencentcloudapi_bge_base_zh-v1.5"
The following information is returned: (The api_key is hidden).
{
"models" : [
  {
       "model_id" : "tencentcloudapi_bge_base_zh-v1.5"
       "task_type" : "text_embedding",
       "service" : "tencent_cloud_ai_search",
       "service_settings" : {
                "model_id" : "bge-base-zh-v1.5"
                "url" : "https://aisearch.test.tencentcloudapi.com",
                "Language" : "zh-CN"
                "region": "ap-guangzhou",
                "action" : "GetTextEmbedding"
                "version": "2025-01-01"
         },
        "task_settings" :{ }
      }
  ]
}
Create a pipeline based on the atomic service model.
curl -X PUT "localhost:9200/_ingest/pipeline/tencentcloudapi_bge_base_zh-v1.5_embeddings?pretty" -H 'Content-Type: application/json' -d'
{
  "processors": [
    {
      "inference": {
        "model_id": "tencentcloudapi_bge_base_zh-v1.5", 
        "input_output": { 
          "input_field": "content",
          "output_field": "content_embedding"
        }
      }
    }
  ]
}
'
Call atomic services for bulk writes.
curl -u "elastic:changeme" -H "Content-Type: application/x-ndjson; charset=UTF-8" -XPOST "127.0.0.1:9200/vector_index/_bulk?pipeline=tencentcloudapi_bge_base_zh-v1.5_embeddings" -d '
{ "index" : {} }
{ "title" : "value1","content": "good day" }
'
Call the atomic service to perform vector similarity search
curl -H "Content-Type: application/x-ndjson; charset=UTF-8" -XPOST "127.0.0.1:9200/vector_index/_search" -d '{
  "knn": {
    "field": "content_embedding",
    "query_vector_builder": {
      "text_embedding": {
        "model_id": "tencentcloudapi_bge_base_zh-v1.5",
        "model_text": "The original text you need to vectorize"
      }
    },
    "k": 10,
    "num_candidates": 100
  }}'
Rerank API
Create an atomic service rerank model.
(for which keys can be fully overwritten during updates)
 
curl  -H "Content-Type: application/json" -XPUT "http://127.0.0.1:9200/_inference/rerank/tencentcloudapi_bge-reranker-large" -d '{
    "service": "tencent_cloud_ai_search",
    "service_settings": {
         "secret_id": "AKIDI***********************",
        "secret_key": "BQa*************************", 
        "url": "https://es.internal.tencentcloudapi.com",
        "model_id": "bge-reranker-large",
        "region":  "ap-beijing",
        "language":  "zh-CN",
        "action":  "RunRerank",
        "version":  "2025-01-01"
    },
    "task_settings": {
        "top_n": 10,
        "return_documents": true
    }
}'
 
Note 1: Use secret_id and secret_key for the key. Ensure that the TencentCloud API key permission includes ES API calls.
Note 2: The url must start with http/https. No modification is needed by default.
Note 3: For model_id, enter the model name, such as bge-reranker-large. Custom models will be supported later.
Note 4: For region, you can enter ap-beijing. Currently, the atomic service is deployed only in the Beijing region.
Call the atomic service rerank model for reranking.
 
curl  -H "Content-Type: application/json" -XPOST "http://127.0.0.1:9200/_inference/rerank/tencentcloudapi_bge-reranker-large" -d '{
    "query": "China",
    "input": ["United States", "China", "United Kingdom"]
}'
Query and rerank 
integrated request
curl  -H "Content-Type: application/json"  -XGET "http://127.0.0.1:9200/vector_index/_search?pretty" -d '
{  
  "retriever": {                                                      
    "tencent_cloud_ai_reranker": {                           
      "retriever": {
        "standard": {                                                 
          "query": {
            "match_all": {                                                // Query condition
            }
          }
        }
      },
      "model_id": "tencentcloudapi_bge-reranker-large",                           
      "rank_field": "content",                                                                                 
      "rank_text": "nice day",
      "rank_window_size": 10,                                                             
      "min_score": 0.6                                                                                 
    }
  }
}'
 
 
curl  -H "Content-Type: application/json"  -XGET "http://127.0.0.1:9200/vector_index/_search?pretty" -d '
'
Reference: Getting and Updating Inference APIs in Batches (8.16)
Batch capabilities
 
Get inference API
Update inference API
Basic usage: Use the official get and update APIs. Refer to the official documentation above.
Batch capability usage: The batch capability is added through self-development. Refer to the examples below.
Example of getting inference endpoints in batches
curl  -H "Content-Type: application/json" -XGET "http://127.0.0.1:9200/_inference/completion/a*,b*" 
curl  -H "Content-Type: application/json" -XGET "http://127.0.0.1:9200/_inference/a*,b*" 
Example of updating inference endpoints in batches
curl  -H "Content-Type: application/json" -XPUT "http://127.0.0.1:9200/_inference/rerank/a*,b*/_update" -d '{
    "service_settings": {
        "secret_id": "AKI3*************",
        "secret_key": "GZ4*********"
    }
}'
 
Note 1: Use secret_id and secret_key for the key. Ensure that the TencentCloud API key permission includes ES API calls.

Bantuan dan Dukungan

Apakah halaman ini membantu?

Anda juga dapat Menghubungi Penjualan atau Mengirimkan Tiket untuk meminta bantuan.

masukan

tencent cloud

Elasticsearch Service

Calling Atomic Services for Inference

Creating an Inference Endpoint from an Atomic Service

Reference: Overview of Atomic Services

Embedding Service

Rerank Service

Reference: Inference API Call Guide

Embedding API

Rerank API

Reference: Getting and Updating Inference APIs in Batches (8.16)

Bantuan dan Dukungan

Atomic Service	Token Limit	Dimension	Language	Remarks
bge-base-zh-v1.5	512	768	Chinese	Classic BGE model.
KaLM-embedding-multilingual-mini-v1	131072	896	Multilingual	Self-developed embedding model of WeChat based on an autoregressive LLM with high-quality training data, suitable for scenarios involving ultra-long texts, Chinese-English mixed queries, and multilingual document matching.
bge-m3	8194	1024	Multilingual	Classic BGE model.
conan-embedding-v1	512	1792	Chinese	Self-developed by Tencent, this model once ranked first overall on the MTEB leaderboard for Chinese scenarios last year.

Create an atomic service Embedding model in Elasticsearch, for example, named tencentcloudapi_bge_base_zh-v1.5. (for which keys can be fully overwritten during updates)	curl -H "Content-Type: application/json" -XPUT "http://127.0.0.1:9200/_inference/text_embedding/tencentcloudapi_bge_base_zh-v1.5" -d '{ "service": "tencent_cloud_ai_search", "service_settings": { "secret_id": "AKIDI*********************", "secret_key": "BQa***********************", "url": "https://es.internal.tencentcloudapi.com", "model_id": "bge-base-zh-v1.5", "region": "ap-beijing", "language": "zh-CN", "action": "GetTextEmbedding", "version": "2025-01-01" } }' Note 1: Use secret_id and secret_key for the key. Ensure that the TencentCloud API key permission includes ES API calls. Note 2: The url must start with http/https. No modification is needed by default. Note 3: For model_id, enter the model name, such as bge-base-zh-v1.5 or Conan-embedding-v1. Custom models will be supported later. Note 4: For region, you can enter ap-beijing. Currently, the atomic service is deployed only in the Beijing region. Remark 5: Other fields usually do not need to be changed.
Call the atomic service model to perform vectorization separately	curl -H "Content-Type: application/json" -XPOST "http://127.0.0.1:9200/_inference/text_embedding/tencentcloudapi_bge_base_zh-v1.5" -d '{ "input": ["China", "United States", "United Kingdom"] }'
View models	Obtain all models: GET /_inference/_all. Or obtain a single model: curl -XGET "http://127.0.0.1:9200/_inference/text_embedding/tencentcloudapi_bge_base_zh-v1.5" The following information is returned: (The api_key is hidden). { "models" : [ { "model_id" : "tencentcloudapi_bge_base_zh-v1.5" "task_type" : "text_embedding", "service" : "tencent_cloud_ai_search", "service_settings" : { "model_id" : "bge-base-zh-v1.5" "url" : "https://aisearch.test.tencentcloudapi.com", "Language" : "zh-CN" "region": "ap-guangzhou", "action" : "GetTextEmbedding" "version": "2025-01-01" }, "task_settings" :{ } } ] }
Create a pipeline based on the atomic service model.	curl -X PUT "localhost:9200/_ingest/pipeline/tencentcloudapi_bge_base_zh-v1.5_embeddings?pretty" -H 'Content-Type: application/json' -d' { "processors": [ { "inference": { "model_id": "tencentcloudapi_bge_base_zh-v1.5", "input_output": { "input_field": "content", "output_field": "content_embedding" } } } ] } '
Call atomic services for bulk writes.	curl -u "elastic:changeme" -H "Content-Type: application/x-ndjson; charset=UTF-8" -XPOST "127.0.0.1:9200/vector_index/_bulk?pipeline=tencentcloudapi_bge_base_zh-v1.5_embeddings" -d ' { "index" : {} } { "title" : "value1","content": "good day" } '
Call the atomic service to perform vector similarity search	curl -H "Content-Type: application/x-ndjson; charset=UTF-8" -XPOST "127.0.0.1:9200/vector_index/_search" -d '{ "knn": { "field": "content_embedding", "query_vector_builder": { "text_embedding": { "model_id": "tencentcloudapi_bge_base_zh-v1.5", "model_text": "The original text you need to vectorize" } }, "k": 10, "num_candidates": 100 }}'

Create an atomic service rerank model. (for which keys can be fully overwritten during updates)	curl -H "Content-Type: application/json" -XPUT "http://127.0.0.1:9200/_inference/rerank/tencentcloudapi_bge-reranker-large" -d '{ "service": "tencent_cloud_ai_search", "service_settings": { "secret_id": "AKIDI*********************", "secret_key": "BQa***********************", "url": "https://es.internal.tencentcloudapi.com", "model_id": "bge-reranker-large", "region": "ap-beijing", "language": "zh-CN", "action": "RunRerank", "version": "2025-01-01" }, "task_settings": { "top_n": 10, "return_documents": true } }' Note 1: Use secret_id and secret_key for the key. Ensure that the TencentCloud API key permission includes ES API calls. Note 2: The url must start with http/https. No modification is needed by default. Note 3: For model_id, enter the model name, such as bge-reranker-large. Custom models will be supported later. Note 4: For region, you can enter ap-beijing. Currently, the atomic service is deployed only in the Beijing region.
Call the atomic service rerank model for reranking.	curl -H "Content-Type: application/json" -XPOST "http://127.0.0.1:9200/_inference/rerank/tencentcloudapi_bge-reranker-large" -d '{ "query": "China", "input": ["United States", "China", "United Kingdom"] }'
Query and rerank integrated request	curl -H "Content-Type: application/json" -XGET "http://127.0.0.1:9200/vector_index/_search?pretty" -d ' { "retriever": { "tencent_cloud_ai_reranker": { "retriever": { "standard": { "query": { "match_all": { // Query condition } } } }, "model_id": "tencentcloudapi_bge-reranker-large", "rank_field": "content", "rank_text": "nice day", "rank_window_size": 10, "min_score": 0.6 } } }' curl -H "Content-Type: application/json" -XGET "http://127.0.0.1:9200/vector_index/_search?pretty" -d ' '

Batch capabilities	Get inference API Update inference API Basic usage: Use the official get and update APIs. Refer to the official documentation above. Batch capability usage: The batch capability is added through self-development. Refer to the examples below.
Example of getting inference endpoints in batches	curl -H "Content-Type: application/json" -XGET "http://127.0.0.1:9200/_inference/completion/a,b" curl -H "Content-Type: application/json" -XGET "http://127.0.0.1:9200/_inference/a,b"
Example of updating inference endpoints in batches	curl -H "Content-Type: application/json" -XPUT "http://127.0.0.1:9200/_inference/rerank/a,b/_update" -d '{ "service_settings": { "secret_id": "AKI3***********", "secret_key": "GZ4*******" } }' Note 1: Use secret_id and secret_key for the key. Ensure that the TencentCloud API key permission includes ES API calls.