Sold by: Synthefy
Deployed on AWS
Synthefy-Nori is a foundation model for tabular data that replaces the train-and-tune workflow: pass your labeled rows as context and get predictions in a single forward pass - no training, tuning, feature engineering, or cross-validation. It achieves the highest mean R2 of any tabular foundation model across 96 public regression datasets and beats tuned gradient boosting (XGBoost/LightGBM) on 9 of 13 TabArena tasks, at just 6M parameters. It deploys as an Amazon SageMaker endpoint that runs entirely within your own AWS account.
Overview
Synthefy-Nori is an in-context learning foundation model for tabular regression. Instead of building and tuning a separate model for every dataset, you pass your labeled rows (X_train, y_train) as context and the rows you want scored (X_test) in a single request, and predictions come back in one forward pass. No gradient updates, no hyperparameter search, no validation sweep. The model handles preprocessing, high dimensionality, missing values, and skewed targets on its own, so you can hand it raw rows. When your data drifts, you simply send the new rows as context; there is nothing to retrain.
Across 96 regression datasets from three independent sources (TabArena, TALENT, and OpenML-Reg), Synthefy-Nori achieves the highest mean R2 of any tabular foundation model. With zero tuning it beats tuned gradient-boosting baselines (XGBoost and LightGBM, AutoGluon best-quality) on 9 of 13 TabArena tasks. It delivers this at 6M parameters, roughly a tenth the size of comparable tabular foundation models, and returns predictions for typical small-to-mid tables in about a second on a single GPU.
Synthefy-Nori is delivered as an Amazon SageMaker model package that you deploy and run inside your own AWS account, so your data never leaves your environment. It is permissively licensed (Apache 2.0) for commercial use, and its scikit-learn-style fit/predict interface drops into existing pipelines.
Highlights
- Replaces the train-and-tune loop - no training, hyperparameter search, feature engineering, or cross-validation. Pass labeled rows as context and get predictions in one forward pass; when data drifts, just send new rows.
- Best-in-class accuracy at 6M parameters - the highest mean R2 of any tabular foundation model across 96 public regression datasets, and beats tuned gradient boosting on 9 of 13 TabArena tasks.
- Runs in your account - deploys as an Amazon SageMaker endpoint so your data never leaves your AWS environment, with a simple JSON API: send X_train, y_train, and X_test to get one prediction per row.
Details
Sold by
Categories
Delivery method
Latest version
Deployed on AWS
New
Introducing multi-product solutions
You can now purchase comprehensive solutions tailored to use cases and industries.
Features and programs
Financing for AWS Marketplace purchases
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.
Pricing
This product is available free of charge. Free subscriptions have no end date and may be canceled any time.
Additional AWS infrastructure costs may apply. Use the AWS Pricing Calculator to estimate your infrastructure costs.
Dimensions summary
You pay by host-hour (HostHrs) for running model inference, with charges based on the instance type and mode you select. Batch mode covers offline predictions on grouped inputs, while real-time mode handles live requests. Two instance types run in batch mode: ml.g5.xlarge and ml.g6.xlarge. Three instance types run in real-time mode: ml.g5.xlarge, ml.g5.2xlarge, and ml.g6.xlarge. Each dimension bills independently for the hours that instance stays active. Larger instance sizes and additional processing modes give you separate billing lines, so cost scales with the compute you provision.
Top-of-mind questions for buyers
A host-hour is one hour that a single inference instance stays active in your account. You pay for each hour the instance runs, whether in batch or real-time mode. Larger instance sizes and each mode bill on separate lines, so total cost tracks the hours you keep instances provisioned.
Charges accrue for each hour the instance stays active, not per prediction. A real-time endpoint that idles still bills host-hours while running. The hosted service can scale to zero when idle; the first request afterward triggers a warmup of about 60 to 90 seconds before it returns.
Batch mode runs offline predictions on grouped inputs, while real-time mode handles live requests through a running endpoint. Both bill by host-hour on the instance you select. Batch suits scheduled jobs where the instance runs only during processing. Real-time keeps an endpoint active to serve immediate requests.
docs.synthefy.com
Helpful?
Vendor refund policy
Synthefy-Nori is offered free of charge on AWS Marketplace, so there are no software fees to refund. Buyers pay only their own AWS infrastructure costs (for example, SageMaker instance charges), which are billed directly by AWS and are outside Synthefy's control. For any questions, contact support@synthefy.com .
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.
Delivery details
Amazon SageMaker model
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:
Version release notes
This version expands Synthefy-Nori from a single model to three peer regression models served by one SageMaker endpoint:
- nori-6m
- nori-30m
- nori-30m-thinking-medium
Each request selects its model using the required model field. There is no default or primary model.
Real-time inference now uses SageMaker response streaming for every model, allowing requests to run for up to eight minutes. The released Synthefy Python client 6.3.0 supports SageMaker endpoints, signs requests through the standard AWS credential chain, and handles the streamed response.
The package is GPU-only and supports ml.g5.xlarge, ml.g5.2xlarge, and ml.g6.xlarge for real-time inference. The container does not report ready when CUDA is unavailable.
This version was validated against one immutable container image and one SageMaker model package. Marketplace image scanning, GPU validation, all three model invocations, released-client invocation, and resource cleanup completed successfully.
Additional details
Inputs
- Summary
Synthefy-Nori performs in-context tabular regression. Each request contains labeled context rows and the query rows to predict. No separate training job is required.
Content type: application/json
Fields:
model: Required string. Selects nori-6m, nori-30m, or nori-30m-thinking-medium.
X_train: Required two-dimensional array containing labeled context rows. Shape: [context rows][features]. Feature values must be numbers or null.
y_train: Required one-dimensional array containing one numeric regression target for each X_train row.
X_test: Required two-dimensional array containing the rows to predict. X_test must use the same features and column order as X_train.
task: Optional string. The only supported value is regression, which is also the default.
The response contains one prediction for each X_test row, preserving row order.
- Limitations for input type
- Real-time request bodies are limited to 25 MB and must finish within SageMaker's eight-minute streaming window. Only regression is supported. X_train and X_test must be two-dimensional and have the same number of features. Feature values may be numbers or null. Every y_train value must be a finite number, with one value per X_train row. Categorical features must be encoded before invocation. nori-30m-thinking-medium supports point predictions only.
- Input MIME type
- application/json
{"model":"nori-6m","task":"regression","X_train":[[0.0,1.0],[1.0,0.0],[0.5,0.5],[0.2,0.8]],"y_train":[0.1,0.9,0.5,0.3],"X_test":[[0.3,0.7],[0.8,0.2]]}
{"model":"nori-6m","task":"regression","X_train":[[0.0,1.0],[1.0,0.0],[0.5,0.5],[0.2,0.8]],"y_train":[0.1,0.9,0.5,0.3],"X_test":[[0.3,0.7],[0.8,0.2]]}
Input data descriptions
The following table describes supported input data fields for real-time inference and batch transform.
Field name | Description | Constraints | Required |
|---|---|---|---|
model | nori-6m, nori-30m, or nori-30m-thinking-medium. | - | Yes |
X_train | Two-dimensional array of labeled context rows. | - | Yes |
y_train | Numeric target array with one value per X_train row. | - | Yes |
X_test | Two-dimensional query array using the same features as X_train. | - | Yes |
task | Regression task. Default and only supported value: regression. | - | No |
Custom attributes
The following table describes custom attributes for real-time inference endpoints.
Field name | Description | Constraints | Required |
|---|---|---|---|
synthefy-response-stream | Required for real-time calls made directly through the AWS SDK. Set CustomAttributes to synthefy-response-stream=v1. SynthefyNoriClient sets this automatically. | - | No |
Resources
Vendor resources
Support
Vendor support
Support is available by email at support@synthefy.com . We assist with deployment, invocation, and functional questions on a best-effort basis, typically responding within two business days.
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.
Similar products
Foundation model for time series forecasting with contextual metadata.
Customer reviews
No customer reviews yet
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.