> For the complete documentation index, see [llms.txt](https://docs.cloud.deepset.ai/llms.txt)

Last modified: 2026-11-03

# DeepsetNvidiaRanker

Rank documents by their relevance to the query using NVIDIA Triton.

## Basic Information

-   Type: `deepset_cloud_custom_nodes.rankers.nvidia.ranker.DeepsetNvidiaRanker`
-   Components it most often connects to:
    -   Retrievers: `DeepsetNvidiaRanker` can receive documents from a Retriever and then rank them.
    -   `PromptBuilder`: `DeepsetNvidiaRanker` can send the ranked documents to `PromptBuilder`, which adds them to the prompt for the LLM.
    -   Any component that outputs a list of documents or accepts a list of documents as input.

## Inputs

| Parameter | Type | Default | Description |
| --- | --- | --- | --- |
| query | str |  | The input query to compare the documents to. |
| documents | List[Document] |  | A list of documents to be ranked. |
| top_k | int | None | None |
| scale_score | bool | None | None |
| calibration_factor | float | None | None |
| score_threshold | float | None | None |

## Outputs

| Parameter | Type | Default | Description |
| --- | --- | --- | --- |
| documents | List[Document] |  | A list of documents closest to the query, sorted from most similar to least similar. |

## Overview

`DeepsetNvidiaRanker` uses the [NVIDIA Triton Inference Server](https://developer.nvidia.com/triton-inference-server) to rank documents by their similarity to the query, assigning similarity scores to each document.

This component runs on optimized hardware and is usable within Haystack Enterprise Platform only, which means it doesn't work if you export it to a local Python file. If you're planning to export, use `SentenceTransformersSimilarityRanker` instead.

> **Default Model**
> 
> The default reranker model for new pipelines is `tomaarsen/Qwen3-Reranker-0.6B-seq-cls`. Older models (`intfloat/simlm-msmarco-reranker`, `BAAI/bge-reranker-v2-m3`, and `svalabs/cross-electra-ms-marco-german-uncased`) are no longer available for new pipelines. If your existing pipeline uses one of these models, it continues to work without any changes, and the model appears labelled as **legacy** in the model list.

The model list on the component card shows only the models loaded in your region. If a saved pipeline uses a model that isn't loaded, the model stays selectable and is tagged *unavailable in this region*. Switch such a pipeline to a model available in your region, and choose an available model when you create new pipelines.

## Usage Example

### Basic Configuration

```yaml
  DeepsetNvidiaRanker:
    type: deepset_cloud_custom_nodes.rankers.nvidia.ranker.DeepsetNvidiaRanker
    init_parameters:
      model: intfloat/simlm-msmarco-reranker
      query_prefix: ''
      document_prefix: ''
      top_k: 10
      embedding_separator: \n
      scale_score: true
      calibration_factor: 1
```

### Using the Component in a Pipeline

This is an example of a query pipeline where `DeepsetNvidiaRanker` receives documents from a `DocumentJoiner`, ranks them, and then returns them as the pipeline output.

Here's the YAML configuration:

```yaml
# haystack-pipeline
components:
  bm25_retriever:
    type: haystack_integrations.components.retrievers.opensearch.bm25_retriever.OpenSearchBM25Retriever
    init_parameters:
      document_store:
        type: haystack_integrations.document_stores.opensearch.document_store.OpenSearchDocumentStore
        init_parameters:
          use_ssl: true
          verify_certs: false
          hosts:
            - ${OPENSEARCH_HOST}
          http_auth:
            - ${OPENSEARCH_USER}
            - ${OPENSEARCH_PASSWORD}
          embedding_dim: 768
          similarity: cosine
      top_k: 20
  embedding_retriever:
    type: haystack_integrations.components.retrievers.opensearch.embedding_retriever.OpenSearchEmbeddingRetriever
    init_parameters:
      document_store:
        type: haystack_integrations.document_stores.opensearch.document_store.OpenSearchDocumentStore
        init_parameters:
          use_ssl: true
          verify_certs: false
          hosts:
            - ${OPENSEARCH_HOST}
          http_auth:
            - ${OPENSEARCH_USER}
            - ${OPENSEARCH_PASSWORD}
          embedding_dim: 768
          similarity: cosine
      top_k: 20
  document_joiner:
    type: haystack.components.joiners.document_joiner.DocumentJoiner
    init_parameters:
      join_mode: concatenate
  DeepsetNvidiaTextEmbedder:
    type: deepset_cloud_custom_nodes.embedders.nvidia.text_embedder.DeepsetNvidiaTextEmbedder
    init_parameters:
      model: intfloat/multilingual-e5-base
      prefix: ''
      suffix: ''
      truncate: null
      normalize_embeddings: false
      timeout: null
      backend_kwargs: null
  DeepsetNvidiaRanker:
    type: deepset_cloud_custom_nodes.rankers.nvidia.ranker.DeepsetNvidiaRanker
    init_parameters:
      model: tomaarsen/Qwen3-Reranker-0.6B-seq-cls
      query_prefix: ''
      document_prefix: ''
      top_k: 10
      score_threshold: null
      meta_fields_to_embed: null
      embedding_separator: \n
      scale_score: true
      calibration_factor: 1
      timeout: null
      backend_kwargs: null
connections:
  - sender: bm25_retriever.documents
    receiver: document_joiner.documents
  - sender: embedding_retriever.documents
    receiver: document_joiner.documents
  - sender: DeepsetNvidiaTextEmbedder.embedding
    receiver: embedding_retriever.query_embedding
  - sender: document_joiner.documents
    receiver: DeepsetNvidiaRanker.documents
max_runs_per_component: 100
metadata: {}
inputs:
  query:
    - bm25_retriever.query
    - DeepsetNvidiaTextEmbedder.text
    - DeepsetNvidiaRanker.query
  filters:
    - bm25_retriever.filters
    - embedding_retriever.filters
outputs:
  documents: DeepsetNvidiaRanker.documents
```

## Parameters

### Init Parameters

These are the parameters you can configure in Pipeline Builder:

| Parameter | Type | Default | Description |
| --- | --- | --- | --- |
| model | DeepsetNVIDIARankingModels | DeepsetNVIDIARankingModels.TOMAARSEN_QWEN3_RERANKER_0_6B_SEQ_CLS | The model to use for ranking. Choose the model from a list on the component card. Legacy models are labelled as **legacy** in the picker and are only available if your pipeline already uses them. |
| query_prefix | str |  | String to prepend to queries |
| document_prefix | str |  | String to prepend to documents |
| top_k | int | 10 | Maximum number of documents to return |
| batch_size | int | 40 | The number of documents to rank at once. |
| score_threshold | float | None | None |
| meta_fields_to_embed | List[str] | None | None |
| embedding_separator | str | \n | Separator used when concatenating metadata fields with the document content. |
| scale_score | bool | True | If `True`, scales raw logit predictions using a Sigmoid activation function. |
| calibration_factor | float | 1 | Calibration factor used with `sigmoid(logits * calibration_factor)`. Only used if `scale_score` is `True`. |
| timeout | float | None | None |
| backend_kwargs | dict | None | None |
