Put candidate passages in relevance order before your application chooses which
context to show or send to a language model. This Software Sushi integration
packages the original BAAI/bge-reranker-v2-m3 model for network-isolated SageMaker
inference, with bounded JSON requests and stable source identifiers.
Typical uses include a second ranking stage after keyword or vector retrieval,
English/Spanish/Portuguese knowledge-base search, and a source review interface
that keeps the original document index available to the application.
The offering provides offline serving and AWS packaging. The original model
weights are also publicly available from BAAI under Apache-2.0; this product does
not provide exclusive weights or a new Software Sushi fine-tune.
A relevance score does not verify an answer. The component does not retrieve
documents, generate answers, detect missing answers, check facts or permissions,
or prove that a passage contains the full requested procedure. It always ranks
the supplied pool, even when none of its passages answers the question. Review
the source text and evaluate the component on your workload before adoption.
Consult the current release status, model card, sources and deployment guide before adoption.
Watch the integration
English walkthrough -
Vertical overview.
The illustrative interface replays recorded AWS outputs on fictional development
examples, including a failure; it is not an included buyer application or an
accuracy sample. English narration is AI-generated with NVIDIA Magpie.
View the released AWS Marketplace offer. Software is USD 0.50 per instance-hour with a 14-day software trial; AWS infrastructure is billed separately, including during the trial.
Highlights
Add a relevance-ranking stage after keyword or vector retrieval and caller access filters
Keep original document IDs and indexes; inspect source text before using ranked context
Bounded JSON and line-split batch requests; complete token checks reject oversize inputs without truncation
AWS Marketplace now accepts line of credit payments through the PNC Vendor Finance program. This program is available to select AWS customers in the US, excluding NV, NC, ND, TN, & VT.
You pay by the hour for the EC2 instance that runs this reranking model, so cost scales with how long each instance stays active. The two dimensions differ by instance type and processing mode. One charges for real-time inference on the ml.g5.xlarge instance, for live, on-demand reranking. The other charges for batch inference on the ml.g6e.xlarge instance, for processing grouped requests. You choose the dimension matching your workload. AWS infrastructure, data transfer, and any model-provider usage are billed separately from the software charges shown here.
Top-of-mind questions for buyers
What do the real-time and batch modes mean for how I am billed?
Both dimensions charge per host-hour of running inference. Real-time mode on the ml.g5.xlarge instance handles live, on-demand reranking requests. Batch mode on the ml.g6e.xlarge instance processes grouped requests together. You pay for the hours the chosen instance runs, regardless of how many requests it handles.
Am I charged when the instance is stopped or idle?
Software charges meter running host-hours, so a stopped instance does not accrue software charges. Keeping an instance active between requests still bills per hour. A fully powered-off instance may still incur separate AWS storage fees for its attached volumes.
What costs are separate from the per-hour software charge shown here?
AWS infrastructure such as EC2, EBS storage, networking, and data transfer is billed separately by AWS. Any third-party model-provider usage may also carry its own charges. The software charge shown covers only the inference software running on the chosen instance type.
softwaresushi.com
Helpful?
Vendor refund policy
If the software did not work as described, email info@softwaresushi.com within 30 days with AWS account ID and endpoint or job name. Software Sushi will refund software charges for that use. AWS infrastructure charges are separate and are not covered. Delete endpoints and stop unwanted jobs to release compute.
How can we make this page better?
Tell us how we can improve this page, or report an issue with this product.
Give us feedbackReport a problem with this product or seller
Legal
Vendor terms and conditions
Upon subscribing to this product, you must acknowledge and agree to the terms and conditions outlined in the vendor's End User License Agreement (EULA).
Content disclaimer
Vendors are responsible for their product descriptions and other product content. AWS does not warrant that vendors' product descriptions or other product content are accurate, complete, reliable, current, or error-free.
An Amazon SageMaker model package is a pre-trained machine learning model ready to use without additional training. Use the model package to create a model on Amazon SageMaker for real-time inference or batch processing. Amazon SageMaker is a fully managed platform for building, training, and deploying machine learning models at scale.
Deploy the model on Amazon SageMaker AI using the following options:
Real-time inference
Deploy the model as an API endpoint for your applications. When you send data to the endpoint, SageMaker processes it and returns results by API response. The endpoint runs continuously until you delete it. You're billed for software and SageMaker infrastructure costs while the endpoint runs. AWS Marketplace models don't support Amazon SageMaker Asynchronous Inference. For more information, see Deploy models for real-time inference .
Batch transform
Deploy the model to process batches of data stored in Amazon Simple Storage Service (Amazon S3). SageMaker runs the job, processes your data, and returns results to Amazon S3. When complete, SageMaker stops the model. You're billed for software and SageMaker infrastructure costs only during the batch job. Duration depends on your model, instance type, and dataset size. AWS Marketplace models don't support Amazon SageMaker Asynchronous Inference. For more information, see Batch transform for inference with Amazon SageMaker AI .
Version release notes
Original-model passage-relevance integration. No Software Sushi adaptation is included. Public human-judged diagnostics are descriptive; native customer accuracy and answer completeness are not established. The immutable image passed actual AWS real-time and batch native portability checks and model-package certification. Final original-relevance scope, sources/limitations and software terms were approved by the product owner.
Additional details
Inputs
Outputs
Usage instructions
Sample notebooks
Inputs
Summary
One UTF-8 object with exactly query, documents and top_k. JSON Lines uses one structured record per invocation.
Limitations for input type
Include 1 to 16 passages with unique IDs matching [A-Za-z0-9_-]{1,64}. Query and passage texts must be nonblank, at most 16,000 Unicode code points each and 32,000 combined. Body limit is 96 KiB. Each complete query/passage pair, including special tokens, must fit 512 tokens. No silent truncation. top_k is an integer from 1 to the supplied count. Callers apply permissions, tenant and current-version filtering before ranking.
Input MIME type
application/json, application/jsonlines
Real-time inference sample input data
{"documents": [{"id": "billing_steps", "text": "Open Settings, choose Billing, edit the address, and save your changes."}, {"id": "delivery", "text": "Delivery tracking appears in your order history."}], "query": "How do I update my billing address?", "top_k": 2}
Batch transform sample input data
{"documents": [{"id": "billing_steps", "text": "Open Settings, choose Billing, edit the address, and save your changes."}, {"id": "delivery", "text": "Delivery tracking appears in your order history."}], "query": "How do I update my billing address?", "top_k": 2}
{"documents": [{"id": "billing_steps", "text": "Abre Configuraci\u00f3n, selecciona Facturaci\u00f3n, edita la direcci\u00f3n y guarda."}, {"id": "delivery", "text": "El seguimiento del env\u00edo aparece en el historial de pedidos."}], "query": "\u00bfC\u00f3mo actualizo mi direcci\u00f3n de facturaci\u00f3n?", "top_k": 2}
{"documents": [{"id": "billing_steps", "text": "Abra Configura\u00e7\u00f5es, selecione Cobran\u00e7a, edite o endere\u00e7o e salve."}, {"id": "delivery", "text": "O rastreamento da entrega aparece no hist\u00f3rico de pedidos."}], "query": "Como atualizo meu endere\u00e7o de cobran\u00e7a?", "top_k": 2}
Input data descriptions
The following table describes supported input data fields for real-time inference and batch transform.
Field name
Description
Constraints
Required
query
Question or search text.
Nonblank UTF-8 text; complete pair must fit 512 actual tokens.
Yes
documents
Caller-authorized passage pool with unique id and text fields.
Contact Software Sushi at info@softwaresushi.com with model version, AWS region, instance type, request ID and a redacted error code. Do not send source documents, credentials or private customer data. No response-time SLA is advertised.
AWS infrastructure support
AWS Support is a one-on-one, fast-response support channel that is staffed 24x7x365 with experienced and technical support engineers. The service helps customers of all sizes and technical abilities to successfully utilize the products and features provided by Amazon Web Services.
General-purpose reranker with instruction following optimized for quality, with improved long-context and code retrieval. Context length: 32K for queries & documents (8K for queries).
Rerank improves search systems by sorting documents based on their semantic similarity to a query. Choose fast, when you need a fast Reranking model. Choose pro, when you need the Reranking model with the best performance.
Be the first to review this product. We've partnered with PeerSpot to gather customer feedback. You can share your experience by writing or recording a review, or scheduling a call with a PeerSpot analyst.