Added a workshop for sagemaker automated inference benchmarking

[dferguson992]

Add: SageMaker AI Automated Benchmarking & Inference Recommendations Workshop

Summary

Adds a new 4-lab workshop that walks users through the full lifecycle of benchmarking and optimizing generative AI model deployments on Amazon SageMaker AI — from initial deployment through automated inference recommendations and cost-per-token comparison.

What's included

File	Description
workshops/inference-benchmarking/README.md	Workshop overview, architecture, prerequisites, and references
lab0/lab0_setup.ipynb	Environment setup — endpoint provisioning, dependencies, region config
lab1/lab1_deploy_and_benchmark.ipynb	Deploy Llama 3.1 8B via JumpStart and run a first benchmark with CreateAIBenchmarkJob
lab2/lab2_benchmarking_nuances.ipynb	Explore how concurrency, token lengths, request rates, and streaming impact TTFT/ITL/throughput
lab3/lab3_inference_recommendations.ipynb	Use CreateAIRecommendationJob to find optimal deployment configs across instance types
lab4/lab4_deploy_and_compare.ipynb	Deploy the recommended config and compare baseline vs. optimized performance + cost
requirements.txt	Python dependencies
utils.py	Shared helpers for metrics parsing, visualization, and comparison tables

Key concepts covered

• SageMaker AI Workload Configs and Benchmark Jobs (NVIDIA AIPerf)
• Automated Inference Recommendations with speculative decoding and kernel tuning
• TTFT, ITL, throughput, and latency percentile analysis
• Cost-per-token comparison across configurations

Prerequisites

• Service quota for ml.g6.12xlarge (or ml.g5.12xlarge)
• HuggingFace token stored as env var (HF_TOKEN) for gated model access
• Available in: us-east-1, us-east-2, us-west-2, ap-southeast-1, ap-northeast-1, eu-central-1, eu-west-1

Testing

• Ran all labs end-to-end in us-west-2 on SageMaker Studio
• Verified cleanup cells properly delete endpoints and inference components
• Estimated total cost for full workshop run: ~$100–145

Estimated lab durations

Lab	Time	Cost
Lab 1	~45 min	~$15–20
Lab 2	~60 min	~$25–35
Lab 3	~90 min	~$40–60
Lab 4	~45 min	~$20–30