Evaluating generative AI fashions with Amazon Nova LLM-as-a-Choose on Amazon SageMaker AI

Evaluating the efficiency of massive language fashions (LLMs) goes past statistical metrics like perplexity or bilingual analysis understudy (BLEU) scores. For many real-world generative AI eventualities, it’s essential to know whether or not a mannequin is producing higher outputs than a baseline or an earlier iteration. That is particularly necessary for purposes akin to summarization, content material era, or clever brokers the place subjective judgments and nuanced correctness play a central position.

As organizations deepen their deployment of those fashions in manufacturing, we’re experiencing an growing demand from prospects who wish to systematically assess mannequin high quality past conventional analysis strategies. Present approaches like accuracy measurements and rule-based evaluations, though useful, can’t totally tackle these nuanced evaluation wants, notably when duties require subjective judgments, contextual understanding, or alignment with particular enterprise necessities. To bridge this hole, LLM-as-a-judge has emerged as a promising strategy, utilizing the reasoning capabilities of LLMs to guage different fashions extra flexibly and at scale.

As we speak, we’re excited to introduce a complete strategy to mannequin analysis by the Amazon Nova LLM-as-a-Choose functionality on Amazon SageMaker AI, a totally managed Amazon Net Providers (AWS) service to construct, prepare, and deploy machine studying (ML) fashions at scale. Amazon Nova LLM-as-a-Choose is designed to ship strong, unbiased assessments of generative AI outputs throughout mannequin households. Nova LLM-as-a-Choose is obtainable as optimized workflows on SageMaker AI, and with it, you can begin evaluating mannequin efficiency towards your particular use circumstances in minutes. Not like many evaluators that exhibit architectural bias, Nova LLM-as-a-Choose has been rigorously validated to stay neutral and has achieved main efficiency on key choose benchmarks whereas intently reflecting human preferences. With its distinctive accuracy and minimal bias, it units a brand new commonplace for credible, production-grade LLM analysis.

Nova LLM-as-a-Choose functionality supplies pairwise comparisons between mannequin iterations, so you may make data-driven selections about mannequin enhancements with confidence.

How Nova LLM-as-a-Choose was skilled

Nova LLM-as-a-Choose was constructed by a multistep coaching course of comprising supervised coaching and reinforcement studying levels that used public datasets annotated with human preferences. For the proprietary part, a number of annotators independently evaluated 1000’s of examples by evaluating pairs of various LLM responses to the identical immediate. To confirm consistency and equity, all annotations underwent rigorous high quality checks, with remaining judgments calibrated to mirror broad human consensus fairly than a person viewpoint.

The coaching knowledge was designed to be each numerous and consultant. Prompts spanned a variety of classes, together with real-world data, creativity, coding, arithmetic, specialised domains, and toxicity, so the mannequin may consider outputs throughout many real-world eventualities. Coaching knowledge included knowledge from over 90 languages and is primarily composed of English, Russian, Chinese language, German, Japanese, and Italian.Importantly, an inside bias examine evaluating over 10,000 human-preference judgments towards 75 third-party fashions confirmed that Amazon Nova LLM-as-a-Choose exhibits solely a 3% mixture bias relative to human annotations. Though it is a vital achievement in lowering systematic bias, we nonetheless suggest occasional spot checks to validate essential comparisons.

Within the following determine, you may see how the Nova LLM-as-a-Choose bias compares to human preferences when evaluating Amazon Nova outputs in comparison with outputs from different fashions. Right here, bias is measured because the distinction between the choose’s desire and human desire throughout 1000’s of examples. A optimistic worth signifies the choose barely favors Amazon Nova fashions, and a damaging worth signifies the alternative. To quantify the reliability of those estimates, 95% confidence intervals had been computed utilizing the usual error for the distinction of proportions, assuming unbiased binomial distributions.

Amazon Nova LLM-as-a-Choose achieves superior efficiency amongst analysis fashions, demonstrating robust alignment with human judgments throughout a variety of duties. For instance, it scores 45% accuracy on JudgeBench (in comparison with 42% for Meta J1 8B) and 68% on PPE (versus 60% for Meta J1 8B). The information from Meta’s J1 8B was pulled from Incentivizing Considering in LLM-as-a-Choose by way of Reinforcement Studying.

These outcomes spotlight the energy of Amazon Nova LLM-as-a-Choose in chatbot-related evaluations, as proven within the PPE benchmark. Our benchmarking follows present greatest practices, reporting reconciled outcomes for positionally swapped responses on JudgeBench, CodeUltraFeedback, Eval Bias, and LLMBar, whereas utilizing single-pass outcomes for PPE.

On this submit, we current a streamlined strategy to implementing Amazon Nova LLM-as-a-Choose evaluations utilizing SageMaker AI, deciphering the ensuing metrics, and making use of this course of to enhance your generative AI purposes.

Overview of the analysis workflow

The analysis course of begins by getting ready a dataset during which every instance features a immediate and two different mannequin outputs. The JSONL format appears like this:

{
   "immediate":"Clarify photosynthesis.",
   "response_A":"Reply A...",
   "response_B":"Reply B..."
}
{
   "immediate":"Summarize the article.",
   "response_A":"Reply A...",
   "response_B":"Reply B..."
}

After getting ready this dataset, you utilize the given SageMaker analysis recipe, which configures the analysis technique, specifies which mannequin to make use of because the choose, and defines the inference settings akin to temperature and top_p.

The analysis runs inside a SageMaker coaching job utilizing pre-built Amazon Nova containers. SageMaker AI provisions compute assets, orchestrates the analysis, and writes the output metrics and visualizations to Amazon Easy Storage Service (Amazon S3).

When it’s full, you may obtain and analyze the outcomes, which embrace desire distributions, win charges, and confidence intervals.

Understanding how Amazon Nova LLM-as-a-Choose works

The Amazon Nova LLM-as-a-Choose makes use of an analysis methodology referred to as binary general desire choose. The binary general desire choose is a technique the place a language mannequin compares two outputs facet by facet and picks the higher one or declares a tie. For every instance, it produces a transparent desire. While you mixture these judgments over many samples, you get metrics like win price and confidence intervals. This strategy makes use of the mannequin’s personal reasoning to evaluate qualities like relevance and readability in an easy, constant approach.

• This choose mannequin is supposed to supply low-latency normal general preferences in conditions the place granular suggestions isn’t mandatory
• The output of this mannequin is one in every of [[A>B]] or [[B>A]]
• Use circumstances for this mannequin are primarily these the place automated, low-latency, normal pairwise preferences are required, akin to automated scoring for checkpoint choice in coaching pipelines

Understanding Amazon Nova LLM-as-a-Choose analysis metrics

When utilizing the Amazon Nova LLM-as-a-Choose framework to match outputs from two language fashions, SageMaker AI produces a complete set of quantitative metrics. You should utilize these metrics to evaluate which mannequin performs higher and the way dependable the analysis is. The outcomes fall into three principal classes: core desire metrics, statistical confidence metrics, and commonplace error metrics.

The core desire metrics report how typically every mannequin’s outputs had been most well-liked by the choose mannequin. The a_scores metric counts the variety of examples the place Mannequin A was favored, and b_scores counts circumstances the place Mannequin B was chosen as higher. The ties metric captures cases during which the choose mannequin rated each responses equally or couldn’t establish a transparent desire. The inference_error metric counts circumstances the place the choose couldn’t generate a legitimate judgment as a result of malformed knowledge or inside errors.

The statistical confidence metrics quantify how probably it’s that the noticed preferences mirror true variations in mannequin high quality fairly than random variation. The winrate experiences the proportion of all legitimate comparisons during which Mannequin B was most well-liked. The lower_rate and upper_rate outline the decrease and higher bounds of the 95% confidence interval for this win price. For instance, a winrate of 0.75 with a confidence interval between 0.60 and 0.85 means that, even accounting for uncertainty, Mannequin B is constantly favored over Mannequin A. The rating subject typically matches the rely of Mannequin B wins however will also be custom-made for extra advanced analysis methods.

The commonplace error metrics present an estimate of the statistical uncertainty in every rely. These embrace a_scores_stderr, b_scores_stderr, ties_stderr, inference_error_stderr, andscore_stderr. Smaller commonplace error values point out extra dependable outcomes. Bigger values can level to a necessity for extra analysis knowledge or extra constant immediate engineering.

Deciphering these metrics requires consideration to each the noticed preferences and the boldness intervals:

• If the winrate is considerably above 0.5 and the boldness interval doesn’t embrace 0.5, Mannequin B is statistically favored over Mannequin A.
• Conversely, if the winrate is beneath 0.5 and the boldness interval is totally beneath 0.5, Mannequin A is most well-liked.
• When the boldness interval overlaps 0.5, the outcomes are inconclusive and additional analysis is beneficial.
• Excessive values in inference_error or massive commonplace errors recommend there might need been points within the analysis course of, akin to inconsistencies in immediate formatting or inadequate pattern dimension.

The next is an instance metrics output from an analysis run:

{
  "a_scores": 16.0,
  "a_scores_stderr": 0.03,
  "b_scores": 10.0,
  "b_scores_stderr": 0.09,
  "ties": 0.0,
  "ties_stderr": 0.0,
  "inference_error": 0.0,
  "inference_error_stderr": 0.0,
  "rating": 10.0,
  "score_stderr": 0.09,
  "winrate": 0.38,
  "lower_rate": 0.23,
  "upper_rate": 0.56
}

On this instance, Mannequin A was most well-liked 16 instances, Mannequin B was most well-liked 10 instances, and there have been no ties or inference errors. The winrate of 0.38 signifies that Mannequin B was most well-liked in 38% of circumstances, with a 95% confidence interval starting from 23% to 56%. As a result of the interval consists of 0.5, this consequence suggests the analysis was inconclusive, and extra knowledge may be wanted to make clear which mannequin performs higher general.

These metrics, routinely generated as a part of the analysis course of, present a rigorous statistical basis for evaluating fashions and making data-driven selections about which one to deploy.

Answer overview

This resolution demonstrates learn how to consider generative AI fashions on Amazon SageMaker AI utilizing the Nova LLM-as-a-Choose functionality. The offered Python code guides you thru your complete workflow.

First, it prepares a dataset by sampling questions from SQuAD and producing candidate responses from Qwen2.5 and Anthropic’s Claude 3.7. These outputs are saved in a JSONL file containing the immediate and each responses.

We accessed Anthropic’s Claude 3.7 Sonnet in Amazon Bedrock utilizing the bedrock-runtime consumer. We accessed Qwen2.5 1.5B utilizing a SageMaker hosted Hugging Face endpoint.

Subsequent, a PyTorch Estimator launches an analysis job utilizing an Amazon Nova LLM-as-a-Choose recipe. The job runs on GPU cases akin to ml.g5.12xlarge and produces analysis metrics, together with win charges, confidence intervals, and desire counts. Outcomes are saved to Amazon S3 for evaluation.

Lastly, a visualization operate renders charts and tables, summarizing which mannequin was most well-liked, how robust the desire was, and the way dependable the estimates are. By this end-to-end strategy, you may assess enhancements, monitor regressions, and make data-driven selections about deploying generative fashions—all with out handbook annotation.

Conditions

It is advisable to full the next conditions earlier than you may run the pocket book:

1. Make the next quota improve requests for SageMaker AI. For this use case, it is advisable request a minimal of 1 g5.12xlarge occasion. On the Service Quotas console, request the next SageMaker AI quotas, 1 G5 cases (g5.12xlarge) for coaching job utilization
2. (Non-obligatory) You’ll be able to create an Amazon SageMaker Studio area (seek advice from Use fast setup for Amazon SageMaker AI) to entry Jupyter notebooks with the previous position. (You should utilize JupyterLab in your native setup, too.)
   • Create an AWS Identification and Entry Administration (IAM) position with managed insurance policies AmazonSageMakerFullAccess, AmazonS3FullAccess, and AmazonBedrockFullAccess to present required entry to SageMaker AI and Amazon Bedrock to run the examples.
   • Assign as belief relationship to your IAM position the next coverage:

{
    "Model": "2012-10-17",
    "Assertion": [
        {
            "Sid": "",
            "Effect": "Allow",
            "Principal": {
                "Service": [
                    "bedrock.amazonaws.com",
                    "sagemaker.amazonaws.com"
                ]
            },
            "Motion": "sts:AssumeRole"
        }
    ]
}

3. Clone the GitHub repository with the belongings for this deployment. This repository consists of a pocket book that references coaching belongings:

git clone https://github.com/aws-samples/amazon-nova-samples.git
cd customization/SageMakerTrainingJobs/Amazon-Nova-LLM-As-A-Choose/

Subsequent, run the pocket book Nova Amazon-Nova-LLM-as-a-Choose-Sagemaker-AI.ipynb to start out utilizing the Amazon Nova LLM-as-a-Choose implementation on Amazon SageMaker AI.

Mannequin setup

To conduct an Amazon Nova LLM-as-a-Choose analysis, it is advisable generate outputs from the candidate fashions you wish to examine. On this undertaking, we used two completely different approaches: deploying a Qwen2.5 1.5B mannequin on Amazon SageMaker and invoking Anthropic’s Claude 3.7 Sonnet mannequin in Amazon Bedrock. First, we deployed Qwen2.5 1.5B, an open-weight multilingual language mannequin, on a devoted SageMaker endpoint. This was achieved by utilizing the HuggingFaceModel deployment interface. To deploy the Qwen2.5 1.5B mannequin, we offered a handy script so that you can invoke:python3 deploy_sm_model.py

When it’s deployed, inference will be carried out utilizing a helper operate wrapping the SageMaker predictor API:

# Initialize the predictor as soon as
predictor = HuggingFacePredictor(endpoint_name="qwen25-")
def generate_with_qwen25(immediate: str, max_tokens: int = 500, temperature: float = 0.9) -> str:
    """
    Sends a immediate to the deployed Qwen2.5 mannequin on SageMaker and returns the generated response.
    Args:
        immediate (str): The enter immediate/query to ship to the mannequin.
        max_tokens (int): Most variety of tokens to generate.
        temperature (float): Sampling temperature for era.
    Returns:
        str: The model-generated textual content.
    """
    response = predictor.predict({
        "inputs": immediate,
        "parameters": {
            "max_new_tokens": max_tokens,
            "temperature": temperature
        }
    })
    return response[0]["generated_text"]
reply = generate_with_qwen25("What's the Grotto at Notre Dame?")
print(reply)

In parallel, we built-in Anthropic’s Claude 3.7 Sonnet mannequin in Amazon Bedrock. Amazon Bedrock supplies a managed API layer for accessing proprietary basis fashions (FMs) with out managing infrastructure. The Claude era operate used the bedrock-runtime AWS SDK for Python (Boto3) consumer, which accepted a consumer immediate and returned the mannequin’s textual content completion:

# Initialize Bedrock consumer as soon as
bedrock = boto3.consumer("bedrock-runtime", region_name="us-east-1")
# (Claude 3.7 Sonnet) mannequin ID by way of Bedrock
MODEL_ID = "us.anthropic.claude-3-7-sonnet-20250219-v1:0"
def generate_with_claude4(immediate: str, max_tokens: int = 512, temperature: float = 0.7, top_p: float = 0.9) -> str:
    """
    Sends a immediate to the Claude 4-tier mannequin by way of Amazon Bedrock and returns the generated response.
    Args:
        immediate (str): The consumer message or enter immediate.
        max_tokens (int): Most variety of tokens to generate.
        temperature (float): Sampling temperature for era.
        top_p (float): Prime-p nucleus sampling.
    Returns:
        str: The textual content content material generated by Claude.
    """
    payload = {
        "anthropic_version": "bedrock-2023-05-31",
        "messages": [{"role": "user", "content": prompt}],
        "max_tokens": max_tokens,
        "temperature": temperature,
        "top_p": top_p
    }
    response = bedrock.invoke_model(
        modelId=MODEL_ID,
        physique=json.dumps(payload),
        contentType="software/json",
        settle for="software/json"
    )
    response_body = json.hundreds(response['body'].learn())
    return response_body["content"][0]["text"]
reply = generate_with_claude4("What's the Grotto at Notre Dame?")
print(reply)

When you could have each features generated and examined, you may transfer on to creating the analysis knowledge for the Nova LLM-as-a-Choose.

Put together the dataset

To create a practical analysis dataset for evaluating the Qwen and Claude fashions, we used the Stanford Query Answering Dataset (SQuAD), a broadly adopted benchmark in pure language understanding distributed below the CC BY-SA 4.0 license. SQuAD consists of 1000’s of crowd-sourced question-answer pairs masking a various vary of Wikipedia articles. By sampling from this dataset, we made positive that our analysis prompts mirrored high-quality, factual question-answering duties consultant of real-world purposes.

We started by loading a small subset of examples to maintain the workflow quick and reproducible. Particularly, we used the Hugging Face datasets library to obtain and cargo the primary 20 examples from the SQuAD coaching cut up:

from datasets import load_dataset
squad = load_dataset("squad", cut up="prepare[:20]")

This command retrieves a slice of the complete dataset, containing 20 entries with structured fields together with context, query, and solutions. To confirm the contents and examine an instance, we printed out a pattern query and its floor fact reply:

print(squad[3]["question"])
print(squad[3]["answers"]["text"][0])

For the analysis set, we chosen the primary six questions from this subset:

questions = [squad[i]["question"] for i in vary(6)]

Generate the Amazon Nova LLM-as-a-Choose analysis dataset

After getting ready a set of analysis questions from SQuAD, we generated outputs from each fashions and assembled them right into a structured dataset for use by the Amazon Nova LLM-as-a-Choose workflow. This dataset serves because the core enter for SageMaker AI analysis recipes. To do that, we iterated over every query immediate and invoked the 2 era features outlined earlier:

• generate_with_qwen25() for completions from the Qwen2.5 mannequin deployed on SageMaker
• generate_with_claude() for completions from Anthropic’s Claude 3.7 Sonnet in Amazon Bedrock

For every immediate, the workflow tried to generate a response from every mannequin. If a era name failed as a result of an API error, timeout, or different difficulty, the system captured the exception and saved a transparent error message indicating the failure. This made positive that the analysis course of may proceed gracefully even within the presence of transient errors:

import json
output_path = "llm_judge.jsonl"
with open(output_path, "w") as f:
    for q in questions:
        strive:
            response_a = generate_with_qwen25(q)
        besides Exception as e:
            response_a = f"[Qwen2.5 generation failed: {e}]"
        
        strive:
            response_b = generate_with_claude4(q)
        besides Exception as e:
            response_b = f"[Claude 3.7 generation failed: {e}]"
        row = {
            "immediate": q,
            "response_A": response_a,
            "response_B": response_b
        }
        f.write(json.dumps(row) + "n")
print(f"JSONL file created at: {output_path}")

This workflow produced a JSON Strains file named llm_judge.jsonl. Every line incorporates a single analysis report structured as follows:

{
  "immediate": "What's the capital of France?",
  "response_A": "The capital of France is Paris.",
  "response_B": "Paris is the capital metropolis of France."
}

Then, add this llm_judge.jsonl to an S3 bucket that you just’ve predefined:

upload_to_s3(
    "llm_judge.jsonl",
    "s3:///datasets/byo-datasets-dev/custom-llm-judge/llm_judge.jsonl"
)

Launching the Nova LLM-as-a-Choose analysis job

After getting ready the dataset and creating the analysis recipe, the ultimate step is to launch the SageMaker coaching job that performs the Amazon Nova LLM-as-a-Choose analysis. On this workflow, the coaching job acts as a totally managed, self-contained course of that hundreds the mannequin, processes the dataset, and generates analysis metrics in your designated Amazon S3 location.

We use the PyTorch estimator class from the SageMaker Python SDK to encapsulate the configuration for the analysis run. The estimator defines the compute assets, the container picture, the analysis recipe, and the output paths for storing outcomes:

estimator = PyTorch(
    output_path=output_s3_uri,
    base_job_name=job_name,
    position=position,
    instance_type=instance_type,
    training_recipe=recipe_path,
    sagemaker_session=sagemaker_session,
    image_uri=image_uri,
    disable_profiler=True,
    debugger_hook_config=False,
)

When the estimator is configured, you provoke the analysis job utilizing the match() methodology. This name submits the job to the SageMaker management aircraft, provisions the compute cluster, and begins processing the analysis dataset:

estimator.match(inputs={"prepare": evalInput})

Outcomes from the Amazon Nova LLM-as-a-Choose analysis job

The next graphic illustrates the outcomes of the Amazon Nova LLM-as-a-Choose analysis job.

To assist practitioners rapidly interpret the end result of a Nova LLM-as-a-Choose analysis, we created a comfort operate that produces a single, complete visualization summarizing key metrics. This operate, plot_nova_judge_results, makes use of Matplotlib and Seaborn to render a picture with six panels, every highlighting a special perspective of the analysis consequence.

This operate takes the analysis metrics dictionary—produced when the analysis job is full—and generates the next visible parts:

• Rating distribution bar chart – Reveals what number of instances Mannequin A was most well-liked, what number of instances Mannequin B was most well-liked, what number of ties occurred, and the way typically the choose failed to supply a call (inference errors). This supplies a direct sense of how decisive the analysis was and whether or not both mannequin is dominating.
• Win price with 95% confidence interval – Plots Mannequin B’s general win price towards Mannequin A, together with an error bar reflecting the decrease and higher bounds of the 95% confidence interval. A vertical reference line at 50% marks the purpose of no desire. If the boldness interval doesn’t cross this line, you may conclude the result’s statistically vital.
• Choice pie chart – Visually shows the proportion of instances Mannequin A, Mannequin B, or neither was most well-liked. This helps rapidly perceive desire distribution among the many legitimate judgments.
• A vs. B rating comparability bar chart – Compares the uncooked counts of preferences for every mannequin facet by facet. A transparent label annotates the margin of distinction to emphasise which mannequin had extra wins.
• Win price gauge – Depicts the win price as a semicircular gauge with a needle pointing to Mannequin B’s efficiency relative to the theoretical 0–100% vary. This intuitive visualization helps nontechnical stakeholders perceive the win price at a look.
• Abstract statistics desk – Compiles numerical metrics—together with whole evaluations, error counts, win price, and confidence intervals—right into a compact, clear desk. This makes it simple to reference the precise numeric values behind the plots.

As a result of the operate outputs a normal Matplotlib determine, you may rapidly save the picture, show it in Jupyter notebooks, or embed it in different documentation.

Clear up

Full the next steps to scrub up your assets:

1. Delete your Qwen 2.5 1.5B Endpoint

   import boto3
   
   # Create a low-level SageMaker service consumer.
   
   sagemaker_client = boto3.consumer('sagemaker', region_name=)
   
   # Delete endpoint
   
   sagemaker_client.delete_endpoint(EndpointName=endpoint_name)

2. When you’re utilizing a SageMaker Studio JupyterLab pocket book, shut down the JupyterLab pocket book occasion.

How you should use this analysis framework

The Amazon Nova LLM-as-a-Choose workflow gives a dependable, repeatable approach to match two language fashions by yourself knowledge. You’ll be able to combine this into mannequin choice pipelines to determine which model performs greatest, or you may schedule it as a part of steady analysis to catch regressions over time.

For groups constructing agentic or domain-specific programs, this strategy supplies richer perception than automated metrics alone. As a result of your complete course of runs on SageMaker coaching jobs, it scales rapidly and produces clear visible experiences that may be shared with stakeholders.

Conclusion

This submit demonstrates how Nova LLM-as-a-Choose—a specialised analysis mannequin obtainable by Amazon SageMaker AI—can be utilized to systematically measure the relative efficiency of generative AI programs. The walkthrough exhibits learn how to put together analysis datasets, launch SageMaker AI coaching jobs with Nova LLM-as-a-Choose recipes, and interpret the ensuing metrics, together with win charges and desire distributions. The totally managed SageMaker AI resolution simplifies this course of, so you may run scalable, repeatable mannequin evaluations that align with human preferences.

We suggest beginning your LLM analysis journey by exploring the official Amazon Nova documentation and examples. The AWS AI/ML neighborhood gives intensive assets, together with workshops and technical steering, to assist your implementation journey.

To be taught extra, go to:

In regards to the authors

Surya Kari is a Senior Generative AI Knowledge Scientist at AWS, specializing in creating options leveraging state-of-the-art basis fashions. He has intensive expertise working with superior language fashions together with DeepSeek-R1, the Llama household, and Qwen, specializing in their fine-tuning and optimization. His experience extends to implementing environment friendly coaching pipelines and deployment methods utilizing AWS SageMaker. He collaborates with prospects to design and implement generative AI options, serving to them navigate mannequin choice, fine-tuning approaches, and deployment methods to attain optimum efficiency for his or her particular use circumstances.

Joel Carlson is a Senior Utilized Scientist on the Amazon AGI basis modeling group. He primarily works on creating novel approaches for enhancing the LLM-as-a-Choose functionality of the Nova household of fashions.

Saurabh Sahu is an utilized scientist within the Amazon AGI Basis modeling group. He obtained his PhD in Electrical Engineering from College of Maryland School Park in 2019. He has a background in multi-modal machine studying engaged on speech recognition, sentiment evaluation and audio/video understanding. At present, his work focuses on creating recipes to enhance the efficiency of LLM-as-a-judge fashions for varied duties.

Morteza Ziyadi is an Utilized Science Supervisor at Amazon AGI, the place he leads a number of tasks on post-training recipes and (Multimodal) massive language fashions within the Amazon AGI Basis modeling group. Earlier than becoming a member of Amazon AGI, he spent 4 years at Microsoft Cloud and AI, the place he led tasks targeted on creating pure language-to-code era fashions for varied merchandise. He has additionally served as an adjunct college at Northeastern College. He earned his PhD from the College of Southern California (USC) in 2017 and has since been actively concerned as a workshop organizer, and reviewer for quite a few NLP, Pc Imaginative and prescient and machine studying conferences.

Pradeep Natarajan is a Senior Principal Scientist in Amazon AGI Basis modeling group engaged on post-training recipes and Multimodal massive language fashions. He has 20+ years of expertise in creating and launching a number of large-scale machine studying programs. He has a PhD in Pc Science from College of Southern California.

Michael Cai is a Software program Engineer on the Amazon AGI Customization Group supporting the event of analysis options. He obtained his MS in Pc Science from New York College in 2024. In his spare time he enjoys 3d printing and exploring modern tech.