I’m happy to share that Amazon SageMaker now comes with an improved model deployment experience to help you deploy traditional machine learning (ML) models and foundation models (FMs) faster.

As a data scientist or ML practitioner, you can now use the new ModelBuilder class in the SageMaker Python SDK to package models, perform local inference to validate runtime errors, and deploy to SageMaker from your local IDE or SageMaker Studio notebooks.

In SageMaker Studio, new interactive model deployment workflows give you step-by-step guidance on which instance type to choose to find the most optimal endpoint configuration. SageMaker Studio also provides additional interfaces to add models, test inference, and enable auto scaling policies on the deployed endpoints.

New tools in SageMaker Python SDK
The SageMaker Python SDK has been updated with new tools, including ModelBuilder and SchemaBuilder classes that unify the experience of converting models into SageMaker deployable models across ML frameworks and model servers. Model builder automates the model deployment by selecting a compatible SageMaker container and capturing dependencies from your development environment. Schema builder helps to manage serialization and deserialization tasks of model inputs and outputs. You can use the tools to deploy the model in your local development environment to experiment with it, fix any runtime errors, and when ready, transition from local testing to deploy the model on SageMaker with a single line of code.

Let me show you how this works. In the following example, I choose the Falcon-7B model from the Hugging Face model hub. I first deploy the model locally, run a sample inference, perform local benchmarking to find the optimal configuration, and finally deploy the model with the suggested configuration to SageMaker.

First, import the updated SageMaker Python SDK and define a sample model input and output that matches the prompt format for the selected model.

import sagemaker
from sagemaker.serve.builder.model_builder import ModelBuilder
from sagemaker.serve.builder.schema_builder import SchemaBuilder
from sagemaker.serve import Mode

prompt = "Falcons are"
response = "Falcons are small to medium-sized birds of prey related to hawks and eagles."

sample_input = {
    "inputs": prompt,
    "parameters": {"max_new_tokens": 32}
}

sample_output = [{"generated_text": response}]

Then, create a ModelBuilder instance with the Hugging Face model ID, a SchemaBuilder instance with the sample model input and output, define a local model path, and set the mode to LOCAL_CONTAINER to deploy the model locally. The schema builder generates the required functions for serializing and deserializing the model inputs and outputs.

model_builder = ModelBuilder(
    model="tiiuae/falcon-7b",
    schema_builder=SchemaBuilder(sample_input, sample_output),
    model_path="/path/to/falcon-7b",
    mode=Mode.LOCAL_CONTAINER,
	env_vars={"HF_TRUST_REMOTE_CODE": "True"}
)

Next, call build() to convert the PyTorch model into a SageMaker deployable model. The build function generates the required artifacts for the model server, including the inferency.py and serving.properties files.

local_mode_model = model_builder.build()

For FMs, such as Falcon, you can optionally run tune() in local container mode that performs local benchmarking to find the optimal model serving configuration. This includes the tensor parallel degree that specifies the number of GPUs to use if your environment has multiple GPUs available. Once ready, call deploy() to deploy the model in your local development environment.

tuned_model = local_mode_model.tune()
tuned_model.deploy()

Let’s test the model.

updated_sample_input = model_builder.schema_builder.sample_input
print(updated_sample_input)

{'inputs': 'Falcons are',
 'parameters': {'max_new_tokens': 32}}
 
local_tuned_predictor.predict(updated_sample_input)[0]["generated_text"]

In my demo, the model returns the following response:

a type of bird that are known for their sharp talons and powerful beaks. They are also known for their ability to fly at high speeds […]

When you’re ready to deploy the model on SageMaker, call deploy() again, set the mode to SAGEMAKLER_ENDPOINT, and provide an AWS Identity and Access Management (IAM) role with appropriate permissions.

sm_predictor = tuned_model.deploy(
    mode=Mode.SAGEMAKER_ENDPOINT, 
	role="arn:aws:iam::012345678910:role/role_name"
)

This starts deploying your model on a SageMaker endpoint. Once the endpoint is ready, you can run predictions.

new_input = {'inputs': 'Eagles are','parameters': {'max_new_tokens': 32}}
sm_predictor.predict(new_input)[0]["generated_text"])

New SageMaker Studio model deployment experience
You can start the new interactive model deployment workflows by selecting one or more models to deploy from the models landing page or SageMaker JumpStart model details page or by creating a new endpoint from the endpoints details page.

The new workflows help you quickly deploy the selected model(s) with minimal inputs. If you used SageMaker Inference Recommender to benchmark your model, the dropdown will show instance recommendations from that benchmarking.

Without benchmarking your model, the dropdown will display prospective instances that SageMaker predicts could be a good fit based on its own heuristics. For some of the most popular SageMaker JumpStart models, you’ll see an AWS pretested optimal instance type. For other models, you’ll see generally recommended instance types. For example, if I select the Falcon 40B Instruct model in SageMaker JumpStart, I can see the recommended instance types.

However, if I want to optimize the deployment for cost or performance to meet my specific use cases, I could open the Alternate configurations panel to view more options based on data from before benchmarking.

Once deployed, you can test inference or manage auto scaling policies.

Things to know
Here are a couple of important things to know:

Supported ML models and frameworks – At launch, the new SageMaker Python SDK tools support model deployment for XGBoost and PyTorch models. You can deploy FMs by specifying the Hugging Face model ID or SageMaker JumpStart model ID using the SageMaker LMI container or Hugging Face TGI-based container. You can also bring your own container (BYOC) or deploy models using the Triton model server in ONNX format.

Now available
The new set of tools is available today in all AWS Regions where Amazon SageMaker real-time inference is available. There is no cost to use the new set of tools; you pay only for any underlying SageMaker resources that get created.

Learn more

• Amazon SageMaker Model Deployment
• SageMaker Developer Guide

Get started
Explore the new SageMaker model deployment experience in the AWS Management Console today!

— Antje

Today, I’m happy to introduce the ability to use natural language instructions in Amazon SageMaker Canvas to explore, visualize, and transform data for machine learning (ML).

SageMaker Canvas now supports using foundation model- (FM) powered natural language instructions to complement its comprehensive data preparation capabilities for data exploration, analysis, visualization, and transformation. Using natural language instructions, you can now explore and transform your data to build highly accurate ML models. This new capability is powered by Amazon Bedrock.

Data is the foundation for effective machine learning, and transforming raw data to make it suitable for ML model building and generating predictions is key to better insights. Analyzing, transforming, and preparing data to build ML models is often the most time-consuming part of the ML workflow. With SageMaker Canvas, data preparation for ML is seamless and fast with 300+ built-in transforms, analyses, and an in-depth data quality insights report without writing any code. Starting today, the process of data exploration and preparation is faster and simpler in SageMaker Canvas using natural language instructions for exploring, visualizing, and transforming data.

Data preparation tasks are now accelerated through a natural language experience using queries and responses. You can quickly get started with contextual, guided prompts to understand and explore your data.

Say I want to build an ML model to predict house prices Using SageMaker Canvas. First, I need to prepare my housing dataset to build an accurate model. To get started with the new natural language instructions, I open the SageMaker Canvas application, and in the left navigation pane, I choose Data Wrangler. Under the Data tab and from the list of available datasets, I select the canvas-housing-sample.csv as the dataset, then select Create a data flow and choose Create. I see the tabular view of my dataset and an introduction to the new Chat for data prep capability.

I select Chat for data prep, and it displays the chat interface with a set of guided prompts relevant to my dataset. I can use any of these prompts or query the data for something else.

First, I want to understand the quality of my dataset to identify any outliers or anomalies. I ask SageMaker Canvas to generate a data quality report to accomplish this task.

I see there are no major issues with my data. I would now like to visualize the distribution of a couple of features in the data. I ask SageMaker Canvas to plot a chart.

I now want to filter certain rows to transform my data. I ask SageMaker Canvas to remove rows where the population is less than 1,000. Canvas removes those rows, shows me a preview of the transformed data, and also gives me the option to view and update the code that generated the transform.

I am happy with the preview and add the transformed data to my list of data transform steps on the right. SageMaker Canvas adds the step along with the code.

Now that my data is transformed, I can go on to build my ML model to predict house prices and even deploy the model into production using the same visual interface of SageMaker Canvas, without writing a single line of code.

Data preparation has never been easier for ML!

Availability
The new capability in Amazon SageMaker Canvas to explore and transform data using natural language queries is available in all AWS Regions where Amazon SageMaker Canvas and Amazon Bedrock are supported.

Learn more
Amazon SageMaker Canvas product page

Go build!

— Irshad

Today, we are announcing new Amazon SageMaker inference capabilities that can help you optimize deployment costs and reduce latency. With the new inference capabilities, you can deploy one or more foundation models (FMs) on the same SageMaker endpoint and control how many accelerators and how much memory is reserved for each FM. This helps to improve resource utilization, reduce model deployment costs on average by 50 percent, and lets you scale endpoints together with your use cases.

For each FM, you can define separate scaling policies to adapt to model usage patterns while further optimizing infrastructure costs. In addition, SageMaker actively monitors the instances that are processing inference requests and intelligently routes requests based on which instances are available, helping to achieve on average 20 percent lower inference latency.

Key components
The new inference capabilities build upon SageMaker real-time inference endpoints. As before, you create the SageMaker endpoint with an endpoint configuration that defines the instance type and initial instance count for the endpoint. The model is configured in a new construct, an inference component. Here, you specify the number of accelerators and amount of memory you want to allocate to each copy of a model, together with the model artifacts, container image, and number of model copies to deploy.

Let me show you how this works.

New inference capabilities in action
You can start using the new inference capabilities from SageMaker Studio, the SageMaker Python SDK, and the AWS SDKs and AWS Command Line Interface (AWS CLI). They are also supported by AWS CloudFormation.

For this demo, I use the AWS SDK for Python (Boto3) to deploy a copy of the Dolly v2 7B model and a copy of the FLAN-T5 XXL model from the Hugging Face model hub on a SageMaker real-time endpoint using the new inference capabilities.

Create a SageMaker endpoint configuration

import boto3
import sagemaker

role = sagemaker.get_execution_role()
sm_client = boto3.client(service_name="sagemaker")

sm_client.create_endpoint_config(
    EndpointConfigName=endpoint_config_name,
    ExecutionRoleArn=role,
    ProductionVariants=[{
        "VariantName": "AllTraffic",
        "InstanceType": "ml.g5.12xlarge",
        "InitialInstanceCount": 1,
		"RoutingConfig": {
            "RoutingStrategy": "LEAST_OUTSTANDING_REQUESTS"
        }
    }]
)

Create the SageMaker endpoint

sm_client.create_endpoint(
    EndpointName=endpoint_name,
    EndpointConfigName=endpoint_config_name,
)

Before you can create the inference component, you need to create a SageMaker-compatible model and specify a container image to use. For both models, I use the Hugging Face LLM Inference Container for Amazon SageMaker. These deep learning containers (DLCs) include the necessary components, libraries, and drivers to host large models on SageMaker.

Prepare the Dolly v2 model

from sagemaker.huggingface import get_huggingface_llm_image_uri

# Retrieve the container image URI
hf_inference_dlc = get_huggingface_llm_image_uri(
  "huggingface",
  version="0.9.3"
)

# Configure model container
dolly7b = {
    'Image': hf_inference_dlc,
    'Environment': {
        'HF_MODEL_ID':'databricks/dolly-v2-7b',
        'HF_TASK':'text-generation',
    }
}

# Create SageMaker Model
sagemaker_client.create_model(
    ModelName        = "dolly-v2-7b",
    ExecutionRoleArn = role,
    Containers       = [dolly7b]
)

Prepare the FLAN-T5 XXL model

# Configure model container
flant5xxlmodel = {
    'Image': hf_inference_dlc,
    'Environment': {
        'HF_MODEL_ID':'google/flan-t5-xxl',
        'HF_TASK':'text-generation',
    }
}

# Create SageMaker Model
sagemaker_client.create_model(
    ModelName        = "flan-t5-xxl",
    ExecutionRoleArn = role,
    Containers       = [flant5xxlmodel]
)

Now, you’re ready to create the inference component.

Create an inference component for each model
Specify an inference component for each model you want to deploy on the endpoint. Inference components let you specify the SageMaker-compatible model and the compute and memory resources you want to allocate. For CPU workloads, define the number of cores to allocate. For accelerator workloads, define the number of accelerators. RuntimeConfig defines the number of model copies you want to deploy.

# Inference compoonent for Dolly v2 7B
sm_client.create_inference_component(
    InferenceComponentName="IC-dolly-v2-7b",
    EndpointName=endpoint_name,
    VariantName=variant_name,
    Specification={
        "ModelName": "dolly-v2-7b",
        "ComputeResourceRequirements": {
		    "NumberOfAcceleratorDevicesRequired": 2, 
			"NumberOfCpuCoresRequired": 2, 
			"MinMemoryRequiredInMb": 1024
	    }
    },
    RuntimeConfig={"CopyCount": 1},
)

# Inference component for FLAN-T5 XXL
sm_client.create_inference_component(
    InferenceComponentName="IC-flan-t5-xxl",
    EndpointName=endpoint_name,
    VariantName=variant_name,
    Specification={
        "ModelName": "flan-t5-xxl",
        "ComputeResourceRequirements": {
		    "NumberOfAcceleratorDevicesRequired": 2, 
			"NumberOfCpuCoresRequired": 1, 
			"MinMemoryRequiredInMb": 1024
	    }
    },
    RuntimeConfig={"CopyCount": 1},
)

Once the inference components have successfully deployed, you can invoke the models.

Run inference
To invoke a model on the endpoint, specify the corresponding inference component.

import json
sm_runtime_client = boto3.client(service_name="sagemaker-runtime")
payload = {"inputs": "Why is California a great place to live?"}

response_dolly = sm_runtime_client.invoke_endpoint(
    EndpointName=endpoint_name,
    InferenceComponentName = "IC-dolly-v2-7b",
    ContentType="application/json",
    Accept="application/json",
    Body=json.dumps(payload),
)

response_flant5 = sm_runtime_client.invoke_endpoint(
    EndpointName=endpoint_name,
    InferenceComponentName = "IC-flan-t5-xxl",
    ContentType="application/json",
    Accept="application/json",
    Body=json.dumps(payload),
)

result_dolly = json.loads(response_dolly['Body'].read().decode())
result_flant5 = json.loads(response_flant5['Body'].read().decode())

Next, you can define separate scaling policies for each model by registering the scaling target and applying the scaling policy to the inference component. Check out the SageMaker Developer Guide for detailed instructions.

The new inference capabilities provide per-model CloudWatch metrics and CloudWatch Logs and can be used with any SageMaker-compatible container image across SageMaker CPU- and GPU-based compute instances. Given support by the container image, you can also use response streaming.

Now available
The new Amazon SageMaker inference capabilities are available today in AWS Regions US East (Ohio, N. Virginia), US West (Oregon), Asia Pacific (Jakarta, Mumbai, Seoul, Singapore, Sydney, Tokyo), Canada (Central), Europe (Frankfurt, Ireland, London, Stockholm), Middle East (UAE), and South America (São Paulo). For pricing details, visit Amazon SageMaker Pricing. To learn more, visit Amazon SageMaker.

Get started
Log in to the AWS Management Console and deploy your FMs using the new SageMaker inference capabilities today!

— Antje

Today, I’m excited to introduce a new capability in Amazon SageMaker Canvas to use foundation models (FMs) from Amazon Bedrock and Amazon SageMaker Jumpstart through a no-code experience. This new capability makes it easier for you to evaluate and generate responses from FMs for your specific use case with high accuracy.

Every business has its own set of unique domain-specific vocabulary that generic models are not trained to understand or respond to. The new capability in Amazon SageMaker Canvas bridges this gap effectively. SageMaker Canvas trains the models for you so you don’t need to write any code using our company data so that the model output reflects your business domain and use case such as completing a marketing analysis. For the fine-tuning process, SageMaker Canvas creates a new custom model in your account, and the data used for fine-tuning is not used to train the original FM, ensuring the privacy of your data.

Earlier this year, we expanded support for ready-to-use models in Amazon SageMaker Canvas to include foundation models (FMs). This allows you to access, evaluate, and query FMs such as Claude 2, Amazon Titan, and Jurassic-2 (powered by Amazon Bedrock), as well as publicly available models such as Falcon and MPT (powered by Amazon SageMaker JumpStart) through a no-code interface. Extending this experience, we enabled the ability to query the FMs to generate insights from a set of documents in your own enterprise document index, such as Amazon Kendra. While it is valuable to query FMs, customers want to build FMs that generate responses and insights for their use cases. Starting today, a new capability to build FMs addresses this need to generate custom responses.

To get started, I open the SageMaker Canvas application and in the left navigation pane, I choose My models. I select the New model button, select Fine-tune foundation model, and select Create.

I select the training dataset and can choose up to three models to tune. I choose the input column with the prompt text and the output column with the desired output text. Then, I initiate the fine-tuning process by selecting Fine-tune.

Once the fine-tuning process is completed, SageMaker Canvas gives me an analysis of the fine-tuned model with different metrics such as perplexity and loss curves, training loss, validation loss, and more. Additionally, SageMaker Canvas provides a model leaderboard that gives me the ability to measure and compare metrics around model quality for the generated models.

Now, I am ready to test the model and compare responses with the original base model. To test, I select Test in Ready-to-use models from the Analyze page. The fine-tuned model is automatically deployed and is now available for me to chat and compare responses.

Now, I am ready to generate and evaluate insights specific to my use case. The icing on the cake was to achieve this without writing a single line of code.

Learn more

• Amazon SageMaker Canvas product page
• Amazon SageMaker Canvas developer guide

Go build!

— Irshad

PS: Writing a blog post at AWS is always a team effort, even when you see only one name under the post title. In this case, I want to thank Shyam Srinivasan for his technical assistance.

You can use the new OR1 instances to create Amazon OpenSearch Service clusters that use Amazon Simple Storage Service (Amazon S3) for primary storage. You can ingest, store, index, and access just about any imaginable amount of data, while also enjoying a 30% price/performance improvement over existing instance types, eleven nines of data durability, and a zero-time Recovery Point Objective (RPO). You can use this to perform interactive log analytics, monitor application in real time, and more.

New OR1 Instances
These benefits are all made possible by the new OR1 instances, which are available in eight sizes and used for the data nodes of the cluster:

To choose a suitable instance size, read Sizing Amazon OpenSearch Service domains.

The Amazon Elastic Block Store (Amazon EBS) volumes are used for primary storage, with data copied synchronously to S3 as it arrives. The data in S3 is used to create replicas and to rehydrate EBS after shards are moved between instances as a result of a node failure or a routine rebalancing operation. This is made possible by the remote-backed storage and segment replication features that were recently released for OpenSearch.

Creating a Domain
To create a domain I open the Amazon OpenSearch Service Console, select Managed clusters, and click Create domain:

I enter a name for my domain (my-domain), select Standard create, and use the Production template:

Then I choose the Domain with standby deployment option. This option will create active data nodes in two Availability Zones and a standby one in a third. I also choose the latest engine version:

Then I select the OR1 instance family and (for my use case) configure 500 GiB of EBS storage per data node:

I set the other settings as needed, and click Create to proceed:

I take a quick lunch break and when i come back my domain is ready:

Things to Know
Here are a couple of things to know about this new storage option:

Engine Versions – Amazon OpenSearch Service engines version 2.11 and above support OR1 instances.

Regions – The OR1 instance family is available for use with OpenSearch in the US East (Ohio, N. Virginia), US West (N. California, Oregon), Asia Pacific (Mumbai, Singapore, Sydney, Tokyo), and Europe (Frankfurt, Ireland, Spain, Stockholm) AWS Regions.

Pricing – You pay On-Demand or Reserved prices for data nodes, and you also pay for EBS storage. See the Amazon OpenSearch Service Pricing page for more information.

— Jeff;

I’m happy to share that Amazon SageMaker Clarify now supports foundation model (FM) evaluation (preview). As a data scientist or machine learning (ML) engineer, you can now use SageMaker Clarify to evaluate, compare, and select FMs in minutes based on metrics such as accuracy, robustness, creativity, factual knowledge, bias, and toxicity. This new capability adds to SageMaker Clarify’s existing ability to detect bias in ML data and models and explain model predictions.

The new capability provides both automatic and human-in-the-loop evaluations for large language models (LLMs) anywhere, including LLMs available in SageMaker JumpStart, as well as models trained and hosted outside of AWS. This removes the heavy lifting of finding the right model evaluation tools and integrating them into your development environment. It also simplifies the complexity of trying to adopt academic benchmarks to your generative artificial intelligence (AI) use case.

Evaluate FMs with SageMaker Clarify
With SageMaker Clarify, you now have a single place to evaluate and compare any LLM based on predefined criteria during model selection and throughout the model customization workflow. In addition to automatic evaluation, you can also use the human-in-the-loop capabilities to set up human reviews for more subjective criteria, such as helpfulness, creative intent, and style, by using your own workforce or managed workforce from SageMaker Ground Truth.

To get started with model evaluations, you can use curated prompt datasets that are purpose-built for common LLM tasks, including open-ended text generation, text summarization, question answering (Q&A), and classification. You can also extend the model evaluation with your own custom prompt datasets and metrics for your specific use case. Human-in-the-loop evaluations can be used for any task and evaluation metric. After each evaluation job, you receive an evaluation report that summarizes the results in natural language and includes visualizations and examples. You can download all metrics and reports and also integrate model evaluations into SageMaker MLOps workflows.

In SageMaker Studio, you can find Model evaluation under Jobs in the left menu. You can also select Evaluate directly from the model details page of any LLM in SageMaker JumpStart.

Select Evaluate a model to set up the evaluation job. The UI wizard will guide you through the selection of automatic or human evaluation, model(s), relevant tasks, metrics, prompt datasets, and review teams.

Once the model evaluation job is complete, you can view the results in the evaluation report.

In addition to the UI, you can also start with example Jupyter notebooks that walk you through step-by-step instructions on how to programmatically run model evaluation in SageMaker.

Evaluate models anywhere with the FMEval open source library
To run model evaluation anywhere, including models trained and hosted outside of AWS, use the FMEval open source library. The following example demonstrates how to use the library to evaluate a custom model by extending the ModelRunner class.

For this demo, I choose GPT-2 from the Hugging Face model hub and define a custom HFModelConfig and HuggingFaceCausalLLMModelRunner class that works with causal decoder-only models from the Hugging Face model hub such as GPT-2. The example is also available in the FMEval GitHub repo.

!pip install fmeval

# ModelRunners invoke FMs
from amazon_fmeval.model_runners.model_runner import ModelRunner

# Additional imports for custom model
import warnings
from dataclasses import dataclass
from typing import Tuple, Optional
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

@dataclass
class HFModelConfig:
    model_name: str
    max_new_tokens: int
    normalize_probabilities: bool = False
    seed: int = 0
    remove_prompt_from_generated_text: bool = True

class HuggingFaceCausalLLMModelRunner(ModelRunner):
    def __init__(self, model_config: HFModelConfig):
        self.config = model_config
        self.model = AutoModelForCausalLM.from_pretrained(self.config.model_name)
        self.tokenizer = AutoTokenizer.from_pretrained(self.config.model_name)

    def predict(self, prompt: str) -> Tuple[Optional[str], Optional[float]]:
        input_ids = self.tokenizer(prompt, return_tensors="pt").to(self.model.device)
        generations = self.model.generate(
            **input_ids,
            max_new_tokens=self.config.max_new_tokens,
            pad_token_id=self.tokenizer.eos_token_id,
        )
        generation_contains_input = (
            input_ids["input_ids"][0] == generations[0][: input_ids["input_ids"].shape[1]]
        ).all()
        if self.config.remove_prompt_from_generated_text and not generation_contains_input:
            warnings.warn(
                "Your model does not return the prompt as part of its generations. "
                "`remove_prompt_from_generated_text` does nothing."
            )
        if self.config.remove_prompt_from_generated_text and generation_contains_input:
            output = self.tokenizer.batch_decode(generations[:, input_ids["input_ids"].shape[1] :])[0]
        else:
            output = self.tokenizer.batch_decode(generations, skip_special_tokens=True)[0]

        with torch.inference_mode():
            input_ids = self.tokenizer(self.tokenizer.bos_token + prompt, return_tensors="pt")["input_ids"]
            model_output = self.model(input_ids, labels=input_ids)
            probability = -model_output[0].item()

        return output, probability

Next, create an instance of HFModelConfig and HuggingFaceCausalLLMModelRunner with the model information.

hf_config = HFModelConfig(model_name="gpt2", max_new_tokens=32)
model = HuggingFaceCausalLLMModelRunner(model_config=hf_config)

Then, select and configure the evaluation algorithm.

# Let's evaluate the FM for FactualKnowledge
from amazon_fmeval.fmeval import get_eval_algorithm
from amazon_fmeval.eval_algorithms.factual_knowledge import FactualKnowledgeConfig

eval_algorithm_config = FactualKnowledgeConfig("<OR>")
eval_algorithm = get_eval_algorithm("factual_knowledge", eval_algorithm_config)

Let’s first test with one sample. The evaluation score is the percentage of factually correct responses.

model_output = model.predict("London is the capital of")[0]
print(model_output)

eval_algo.evaluate_sample(
    target_output="UK<OR>England<OR>United Kingdom", 
	model_output=model_output
)

the UK, and the UK is the largest producer of food in the world.

The UK is the world's largest producer of food in the world.
[EvalScore(name='factual_knowledge', value=1)]

Although it’s not a perfect response, it includes “UK.”

Next, you can evaluate the FM using built-in datasets or define your custom dataset. If you want to use a custom evaluation dataset, create an instance of DataConfig:

config = DataConfig(
    dataset_name="my_custom_dataset",
    dataset_uri="dataset.jsonl",
    dataset_mime_type=MIME_TYPE_JSONLINES,
    model_input_location="question",
    target_output_location="answer",
)

eval_output = eval_algorithm.evaluate(
    model=model, 
    dataset_config=config, 
    prompt_template="$feature", #$feature is replaced by the input value in the dataset 
    save=True
)

The evaluation results will return a combined evaluation score across the dataset and detailed results for each model input stored in a local output path.

Join the preview
FM evaluation with Amazon SageMaker Clarify is available today in public preview in AWS Regions US East (Ohio), US East (N. Virginia), US West (Oregon), Asia Pacific (Singapore), Asia Pacific (Tokyo), Europe (Frankfurt), and Europe (Ireland). The FMEval open source library] is available on GitHub. To learn more, visit Amazon SageMaker Clarify.

Get started
Log in to the AWS Management Console and start evaluating your FMs with SageMaker Clarify today!

— Antje

Amazon Redshift puts artificial intelligence (AI) at your service to optimize efficiencies and make you more productive with two new capabilities that we are launching in preview today.

First, Amazon Redshift Serverless becomes smarter. It scales capacity proactively and automatically along dimensions such as the complexity of your queries, their frequency, the size of the dataset, and so on to deliver tailored performance optimizations. This allows you to spend less time tuning your data warehouse instances and more time getting value from your data.

Second, Amazon Q generative SQL in Amazon Redshift Query Editor generates SQL recommendations from natural language prompts. This helps you to be more productive in extracting insights from your data.

Let’s start with Amazon Redshift Serverless
When you use Amazon Redshift Serverless, you can now opt in for a preview of AI-driven scaling and optimizations. When enabled, the system observes and learns from your usage patterns, such as the concurrent number of queries, their complexity, and the time it takes to run them. Then, it automatically optimizes your serverless endpoint to meet your price performance target. Based on AWS internal testing, this new capability may give you up to ten times better price performance for variable workloads without any manual intervention.

AI-driven scaling and optimizations eliminate the time and effort to manually resize your workgroup and plan background optimizations based on workload needs. It continually runs automatic optimizations when they are most valuable for better performance, avoiding performance cliffs and time-outs.

This new capability goes beyond the existing self-tuning capabilities of Amazon Redshift Serverless, such as machine learning (ML)-enhanced techniques to adjust your compute, modify the physical schema of the database, create or drop materialized views as needed (the one we manage automatically, not yours), and vacuum tables. This new capability brings more intelligence to decide how to adjust the compute, what background optimizations are required, and when to apply them, and it makes its decisions based on more dimensions. We also orchestrate ML-based optimizations for materialized views, table optimizations, and workload management when your queries need it.

During the preview, you must opt in to enable these AI-driven scaling and optimizations on your workgroups. You configure the system to balance the optimization for price or performance. There is only one slider to adjust in the console.

As usual, you can track resource usage and associated changes through the console, Amazon CloudWatch metrics, and the system table SYS_SERVERLESS_USAGE.

Now, let’s look at Amazon Q generative SQL in Amazon Redshift Query Editor
What if you could use generative AI to help analysts write effective SQL queries more rapidly? This is the new experience we introduce today in Amazon Redshift Query Editor, our web-based SQL editor.

You can now describe the information you want to extract from your data in natural language, and we generate the SQL query recommendations for you. Behind the scenes, Amazon Q generative SQL uses a large language model (LLM) and Amazon Bedrock to generate the SQL query. We use different techniques, such as prompt engineering and Retrieval Augmented Generation (RAG), to query the model based on your context: the database you’re connected to, the schema you’re working on, your query history, and optionally the query history of other users connected to the same endpoint. The system also remembers previous questions. You can ask it to refine a previously generated query.

The SQL generation model uses metadata specific to your data schema to generate relevant queries. For example, it uses the table and column names and the relationship between the tables in your database. In addition, your database administrator can authorize the model to use the query history of all users in your AWS account to generate even more relevant SQL statements. We don’t share your query history with other AWS accounts and we don’t train our generation models with any data coming from your AWS account. We maintain the high level of privacy and security that you expect from us.

Using generated SQL queries helps you to get started when discovering new schemas. It does the heavy lifting of discovering the column names and relationships between tables for you. Senior analysts also benefit from asking what they want in natural language and having the SQL statement automatically generated. They can review the queries and run them directly from their notebook.

Let’s explore a schema and extract information
For this demo, let’s pretend I am a data analyst at a company that sells concert tickets. The database schema and data are available for you to download. My manager asks me to analyze the ticket sales data to send a thank you note with discount coupons to the highest-spending customers in Seattle.

I connect to Amazon Redshift Query Editor and connect the analytic endpoint. I create a new tab for a Notebook (SQL generation is available from notebooks only).

Instead of writing a SQL statement, I open the chat panel and type, “Find the top five users from Seattle who bought the most number of tickets in 2022.” I take the time to verify the generated SQL statement. It seems correct, so I decide to run it. I select Add to notebook and then Run. The query returns the list of the top five buyers in Seattle.

I had no previous knowledge of the data schema, and I did not type a single line of SQL to find the information I needed.

But generative SQL is not limited to a single interaction. I can chat with it to dynamically refine the queries. Here is another example.

I ask “Which state has the most venues?” Generative SQL proposes the following query. The answer is New York, with 49 venues, if you’re curious.

I changed my mind, and I want to know the top three cities with the most venues. I simply rephrase my question: “What about the top three venues?”

I add the query to the notebook and run it. It returns the expected result.

Best practices for prompting
Here are a couple of tips and tricks to get the best results out of your prompts.

Be specific – When asking questions in natural language, be as specific as possible to help the system understand exactly what you need. For example, instead of writing “find the top venues that sold the most tickets,” provide more details like “find the names of the top three venues that sold the most tickets in 2022.” Use consistent entity names like venue, ticket, and location instead of referring to the same entity in different ways, which can confuse the system.

Iterate – Break your complex requests into multiple simple statements that are easier for the system to interpret. Iteratively ask follow-up questions to get more detailed analysis from the system. For example, start by asking, “Which state has the most venues?” Then, based on the response, ask a follow-up question like “Which is the most popular venue from this state?”

Verify – Review the generated SQL before running it to ensure accuracy. If the generated SQL query has errors or does not match your intent, provide instructions to the system on how to correct it instead of rephrasing the entire request. For example, if the query is missing a filter clause on year, write “provide venues from year 2022.”

Availability and pricing
AI-driven scaling and optimizations are in preview in six AWS Regions: US East (Ohio, N. Virginia), US West (Oregon), Asia Pacific (Tokyo), and Europe (Ireland, Stockholm). They come at no additional cost. You pay only for the compute capacity your data warehouse consumes when it is active. Pricing is per Redshift Processing Unit (RPU) per hour. The billing is per second of used capacity. The pricing page for Amazon Redshift has the details.

Amazon Q generative SQL for Amazon Redshift Query Editor is in preview in two AWS Regions today: US East (N. Virginia) and US West (Oregon). There is no charge during the preview period.

These are two examples of how AI helps to optimize performance and increase your productivity, either by automatically adjusting the price-performance ratio of your Amazon Redshift Serverless endpoints or by generating correct SQL statements from natural language prompts.

Previews are essential for us to capture your feedback before we make these capabilities available for all. Experiment with these today and let us know what you think on the re:Post forums or using the feedback button on the bottom left side of the console.

— seb

Today, we’re introducing AWS Clean Rooms ML (preview), a new capability of AWS Clean Rooms that helps you and your partners apply machine learning (ML) models on your collective data without copying or sharing raw data with each other. With this new capability, you can generate predictive insights using ML models while continuing to protect your sensitive data.

During this preview, AWS Clean Rooms ML introduces its first model specialized to help companies create lookalike segments for marketing use cases. With AWS Clean Rooms ML lookalike, you can train your own custom model, and you can invite partners to bring a small sample of their records to collaborate and generate an expanded set of similar records while protecting everyone’s underlying data.

In the coming months, AWS Clean Rooms ML will release a healthcare model. This will be the first of many models that AWS Clean Rooms ML will support next year.

AWS Clean Rooms ML helps you to unlock various opportunities for you to generate insights. For example:

• Airlines can take signals about loyal customers, collaborate with online booking services, and offer promotions to users with similar characteristics.
• Auto lenders and car insurers can identify prospective auto insurance customers who share characteristics with a set of existing lease owners.
• Brands and publishers can model lookalike segments of in-market customers and deliver highly relevant advertising experiences.
• Research institutions and hospital networks can find candidates similar to existing clinical trial participants to accelerate clinical studies (coming soon).

AWS Clean Rooms ML lookalike modeling helps you apply an AWS managed, ready-to-use model that is trained in each collaboration to generate lookalike datasets in a few clicks, saving months of development work to build, train, tune, and deploy your own model.

How to use AWS Clean Rooms ML to generate predictive insights
Today I will show you how to use lookalike modeling in AWS Clean Rooms ML and assume you have already set up a data collaboration with your partner. If you want to learn how to do that, check out the AWS Clean Rooms Now Generally Available — Collaborate with Your Partners without Sharing Raw Data post.

With your collective data in the AWS Clean Rooms collaboration, you can work with your partners to apply ML lookalike modeling to generate a lookalike segment. It works by taking a small sample of representative records from your data, creating a machine learning (ML) model, then applying the particular model to identify an expanded set of similar records from your business partner’s data.

The following screenshot shows the overall workflow for using AWS Clean Rooms ML.

By using AWS Clean Rooms ML, you don’t need to build complex and time-consuming ML models on your own. AWS Clean Rooms ML trains a custom, private ML model, which saves months of your time while still protecting your data.

Eliminating the need to share data
As ML models are natively built within the service, AWS Clean Rooms ML helps you protect your dataset and customer’s information because you don’t need to share your data to build your ML model.

You can specify the training dataset using the AWS Glue Data Catalog table, which contains user-item interactions.

Under Additional columns to train, you can define numerical and categorical data. This is useful if you need to add more features to your dataset, such as the number of seconds spent watching a video, the topic of an article, or the product category of an e-commerce item.

Applying custom-trained AWS-built models
Once you have defined your training dataset, you can now create a lookalike model. A lookalike model is a machine learning model used to find similar profiles in your partner’s dataset without either party having to share their underlying data with each other.

When creating a lookalike model, you need to specify the training dataset. From a single training dataset, you can create many lookalike models. You also have the flexibility to define the date window in your training dataset using Relative range or Absolute range. This is useful when you have data that is constantly updated within AWS Glue, such as articles read by users.

Easy-to-tune ML models
After you create a lookalike model, you need to configure it to use in AWS Clean Rooms collaboration. AWS Clean Rooms ML provides flexible controls that enable you and your partners to tune the results of the applied ML model to garner predictive insights.

On the Configure lookalike model page, you can choose which Lookalike model you want to use and define the Minimum matching seed size you need. This seed size defines the minimum number of profiles in your seed data that overlap with profiles in the training data.

You also have the flexibility to choose whether the partner in your collaboration receives metrics in Metrics to share with other members.

With your lookalike models properly configured, you can now make the ML models available for your partners by associating the configured lookalike model with a collaboration.

Creating lookalike segments
Once the lookalike models have been associated, your partners can now start generating insights by selecting Create lookalike segment and choosing the associated lookalike model for your collaboration.

Here on the Create lookalike segment page, your partners need to provide the Seed profiles. Examples of seed profiles include your top customers or all customers who purchased a specific product. The resulting lookalike segment will contain profiles from the training data that are most similar to the profiles from the seed.

Lastly, your partner will get the Relevance metrics as the result of the lookalike segment using the ML models. At this stage, you can use the Score to make a decision.

Export data and use programmatic API
You also have the option to export the lookalike segment data. Once it’s exported, the data is available in JSON format and you can process this output by integrating with AWS Clean Rooms API and your applications.

Join the preview
AWS Clean Rooms ML is now in preview and available via AWS Clean Rooms in US East (Ohio, N. Virginia), US West (Oregon), Asia Pacific (Seoul, Singapore, Sydney, Tokyo), and Europe (Frankfurt, Ireland, London). Support for additional models is in the works.

Learn how to apply machine learning with your partners without sharing underlying data on the AWS Clean Rooms ML page.

Happy collaborating!
— Donnie

Today we are announcing a preview of Amazon OpenSearch Service zero-ETL integration with Amazon S3, a new way to query operational logs in Amazon S3 and S3-based data lakes without needing to switch between services. You can now analyze infrequently queried data in cloud object stores and simultaneously use the operational analytics and visualization capabilities of OpenSearch Service.

Amazon OpenSearch Service direct queries with Amazon S3 provides a zero-ETL integration to reduce the operational complexity of duplicating data or managing multiple analytics tools by enabling customers to directly query their operational data, reducing costs and time to action. This zero-ETL integration will be configurable within OpenSearch Service, where you can take advantage of various log type templates, including predefined dashboards, and configure data accelerations tailored to that log type. Templates include VPC Flow Logs, Elastic Load Balancing logs, and NGINX logs, and accelerations include skipping indexes, materialized views, and covered indexes.

With direct queries with Amazon S3, you can perform complex queries critical to security forensic and threat analysis that correlate data across multiple data sources, which aids teams in investigating service downtime and security events. After creating an integration, you can start querying their data directly from the OpenSearch Dashboards or OpenSearch API. You can easily audit connections to ensure that they are set up in a scalable, cost-efficient, and secure way.

Getting started with direct queries with Amazon S3
You can easily get started by creating a new Amazon S3 direct query data source for OpenSearch Service through the AWS Management Console or the API. Each new data source uses AWS Glue Data Catalog to manage tables that represent S3 buckets. Once you create a data source, you can configure Amazon S3 tables and data indexing and query data in OpenSearch Dashboards.

1. Create a data source in OpenSearch Service
Before you create a data source, you should have an OpenSearch Service domain with version 2.11 or later and a target Amazon S3 table in AWS Glue Data Catalog with the appropriate IAM permissions. IAM will need access to the desired S3 bucket(s) and read and write access to AWS Glue Data Catalog. To learn more about IAM prerequisites, see Creating a data source in the AWS documentation.

Go to the OpenSearch Service console and choose the domain you want to set up a new data source for. In the domain details page, choose the Connections tab below the general information and see the Direct Query section.

To create a new data source, choose Create, input the name of your new data source, select the data source type as Amazon S3 with AWS Glue Data Catalog, and choose the IAM role for your data source.

Once you create a data source, you can go to the OpenSearch Dashboards of the domain, which you use to configure access control, define tables, set up log type–based dashboards for popular log types, and query your data.

2. Configuring your data source in OpenSearch Dashboards
To configure data source in OpenSearch Dashboards, choose Configure in the console and go to OpenSearch Dashboards. In the left-hand navigation of OpenSearch Dashboards, under Management, choose Data sources. Under Manage data sources, choose the name of the data source you created in the console.

Direct queries from OpenSearch Service to Amazon S3 use Spark tables within AWS Glue Data Catalog. To create a new table you want to direct query, go to the Query Workbench in the Open Search Plugins menu.

Now run as in the following SQL statement to create http_logs table and run MSCK REPAIR TABLE mys3.default.http_logs command to update the metadata in the catalog

CREATE EXTERNAL TABLE IF NOT EXISTS mys3.default.http_logs (
   `@timestamp` TIMESTAMP,
    clientip STRING,
    request STRING, 
    status INT, 
    size INT, 
    year INT, 
    month INT, 
    day INT) 
USING json PARTITIONED BY(year, month, day) OPTIONS (path 's3://mys3/data/http_log/http_logs_partitioned_json_bz2/', compression 'bzip2')

To ensure a fast experience with your data in Amazon S3, you can set up any of three different types of accelerations to index data into OpenSearch Service, such as skipping indexes, materialized views, and covering indexes. To create OpenSearch indexes from external data connections for better performance, choose the Accelerate Table.

• Skipping indexes allow you to index only the metadata of the data stored in Amazon S3. Skipping indexes help quickly identify data stored by narrowing down a specific location of where the data is stored.
• Materialized views enable you to use complex queries such as aggregations, which can be used for querying or powering dashboard visualizations. Materialized views ingest data into OpenSearch Service for anomaly detection or geospatial capabilities.
• Covering indexes will ingest all the data from the specified table column. Covering indexes are the most performant of the three indexing types.

3. Query your data source in OpenSearch Dashboards
After you set up your tables, you can query your data using Discover. You can run a sample SQL query for the http_logs table you created in AWS Glue Data Catalog tables.

To learn more, see Working with Amazon OpenSearch Service direct queries with Amazon S3 in the AWS documentation.

Join the preview
Amazon OpenSearch Service zero-ETL integration with Amazon S3 is now previewed in the AWS US East (Ohio), US East (N. Virginia), US West (Oregon), Asia Pacific (Tokyo), Europe (Frankfurt), and Europe (Ireland) Regions.

OpenSearch Service separately charges for only the compute needed as OpenSearch Compute Units to query your external data as well as maintain indexes in OpenSearch Service. For more information, see Amazon OpenSearch Service Pricing.

Give it a try and send feedback to the AWS re:Post for Amazon OpenSearch Service or through your usual AWS Support contacts.

— Channy

I am happy to announce the general availability of Amazon Neptune Analytics, a new analytics database engine that makes it faster for data scientists and application developers to quickly analyze large amounts of graph data. With Neptune Analytics, you can now quickly load your dataset from Amazon Neptune or your data lake on Amazon Simple Storage Service (Amazon S3), run your analysis tasks in near real time, and optionally terminate your graph afterward.

Graph data enables the representation and analysis of intricate relationships and connections within diverse data domains. Common applications include social networks, where it aids in identifying communities, recommending connections, and analyzing information diffusion. In supply chain management, graphs facilitate efficient route optimization and bottleneck identification. In cybersecurity, they reveal network vulnerabilities and identify patterns of malicious activity. Graph data finds application in knowledge management, financial services, digital advertising, and network security, performing tasks such as identifying money laundering networks in banking transactions and predicting network vulnerabilities.

Since the launch of Neptune in May 2018, thousands of customers have embraced the service for storing their graph data and performing updates and deletion on specific subsets of the graph. However, analyzing data for insights often involves loading the entire graph into memory. For instance, a financial services company aiming to detect fraud may need to load and correlate all historical account transactions.

Performing analyses on extensive graph datasets, such as running common graph algorithms, requires specialized tools. Utilizing separate analytics solutions demands the creation of intricate pipelines to transfer data for processing, which is challenging to operate, time-consuming, and prone to errors. Furthermore, loading large datasets from existing databases or data lakes to a graph analytic solution can take hours or even days.

Neptune Analytics offers a fully managed graph analytics experience. It takes care of the infrastructure heavy lifting, enabling you to concentrate on problem-solving through queries and workflows. Neptune Analytics automatically allocates compute resources according to the graph’s size and quickly loads all the data in memory to run your queries in seconds. Our initial benchmarking shows that Neptune Analytics loads data from Amazon S3 up to 80x faster than existing AWS solutions.

Neptune Analytics supports 5 families of algorithms covering 15 different algorithms, each with multiple variants. For example, we provide algorithms for path-finding, detecting communities (clustering), identifying important data (centrality), and quantifying similarity. Path-finding algorithms are used for use cases such as route planning for supply chain optimization. Centrality algorithms like page rank identify the most influential sellers in a graph. Algorithms like connected components, clustering, and similarity algorithms can be used for fraud-detection use cases to determine whether the connected network is a group of friends or a fraud ring formed by a set of coordinated fraudsters.

Neptune Analytics facilitates the creation of graph applications using openCypher, presently one of the widely adopted graph query languages. Developers, business analysts, and data scientists appreciate openCypher’s SQL-inspired syntax, finding it familiar and structured for composing graph queries.

Let’s see it at work
As we usually do on the AWS News blog, let’s show how it works. For this demo, I first navigate to Neptune in the AWS Management Console. There is a new Analytics section on the left navigation pane. I select Graphs and then Create graph.

On the Create graph page, I enter the details of my graph analytics database engine. I won’t detail each parameter here; their names are self-explanatory.

Pay attention to Allow from public because, the vast majority of the time, you want to keep your graph only available from the boundaries of your VPC. I also create a Private endpoint to allow private access from machines and services inside my account VPC network.

In addition to network access control, users will need proper IAM permissions to access the graph.

Finally, I enable Vector search to perform similarity search using embeddings in the dataset. The dimension of the vector depends on the large language model (LLM) that you use to generate the embedding.

When I am ready, I select Create graph (not shown here).

After a few minutes, my graph is available. Under Connectivity & security, I take note of the Endpoint. This is the DNS name I will use later to access my graph from my applications.

I can also create Replicas. A replica is a warm standby copy of the graph in another Availability Zone. You might decide to create one or more replicas for high availability. By default, we create one replica, and depending on your availability requirements, you can choose not to create replicas.

Business queries on graph data
Now that the Neptune Analytics graph is available, let’s load and analyze data. For the rest of this demo, imagine I’m working in the finance industry.

I have a dataset obtained from the US Securities and Exchange Commission (SEC). This dataset contains the list of positions held by investors that have more than $100 million in assets. Here is a diagram to illustrate the structure of the dataset I use in this demo.

I want to get a better understanding of the positions held by one investment firm (let’s name it “Seb’s Investments LLC”). I wonder what its top five holdings are and who else holds more than $1 billion in the same companies. I am also curious to know what are other investment companies that have a similar portfolio as Seb’s Investments LLC.

To start my analysis, I create a Jupyter notebook in the Neptune section of the AWS Management Console. In the notebook, I first define my analytics endpoint and load the data set from an S3 bucket. It takes only 18 seconds to load 17 million records.

Then, I start to explore the dataset using openCypher queries. I start by defining my parameters:

params = {'name': "Seb's Investments LLC", 'quarter': '2023Q4'}

First, I want to know what the top five holdings are for Seb’s Investments LLC in this quarter and who else holds more than $1 billion in the same companies. In openCypher, it translates to the query hereafter. The $name parameter’s value is “Seb’s Investment LLC” and the $quarter parameter’s value is 2023Q4.

MATCH p=(h:Holder)-->(hq1)-[o:owns]->(holding)
WHERE h.name = $name AND hq1.name = $quarter
WITH DISTINCT holding as holding, o ORDER BY o.value DESC LIMIT 5
MATCH (holding)<-[o2:owns]-(hq2)<--(coholder:Holder)
WHERE hq2.name = '2023Q4'
WITH sum(o2.value) AS totalValue, coholder, holding
WHERE totalValue > 1000000000
RETURN coholder.name, collect(holding.name)

Then, I want to know what the other top five companies are that have similar holdings as “Seb’s Investments LLC.” I use the topKByNode() function to perform a vector search.

MATCH (n:Holder)
WHERE n.name = $name
CALL neptune.algo.vectors.topKByNode(n)
YIELD node, score
WHERE score >0
RETURN node.name LIMIT 5

This query identifies a specific Holder node with the name “Seb’s Investments LLC.” Then, it utilizes the Neptune Analytics custom vector similarity search algorithm on the embedding property of the Holder node to find other nodes in the graph that are similar. The results are filtered to include only those with a positive similarity score, and the query finally returns the names of up to five related nodes.

Pricing and availability
Neptune Analytics is available today in seven AWS Regions: US East (Ohio, N. Virginia), US West (Oregon), Asia Pacific (Singapore, Tokyo), and Europe (Frankfurt, Ireland).

AWS charges for the usage on a pay-as-you-go basis, with no recurring subscriptions or one-time setup fees.

Pricing is based on configurations of memory-optimized Neptune capacity units (m-NCU). Each m-NCU corresponds to one hour of compute and networking capacity and 1 GiB of memory. You can choose configurations starting with 128 m-NCUs and up to 4096 m-NCUs. In addition to m-NCU, storage charges apply for graph snapshots.

I invite you to read the Neptune pricing page for more details

Neptune Analytics is a new analytics database engine to analyze large graph datasets. It helps you discover insights faster for use cases such as fraud detection and prevention, digital advertising, cybersecurity, transportation logistics, and bioinformatics.

Get started
Log in to the AWS Management Console to give Neptune Analytics a try.

— seb