- Who
- AWS
- What
- demonstrates a workflow using SageMaker HyperPod Data Capture to analyze inference traffic and fine-tune a model
- Quote
In this post, we walk through one complete example on a single SageMaker HyperPod cluster.
What happened
AWS demonstrates how to use SageMaker HyperPod Data Capture to record inference traffic, analyze it with Athena, create a SQL labeling policy, and fine-tune a support router on the same cluster. In the synthetic walkthrough, a 60-second LoRA fine-tune raised held-out accuracy from 83.9% to 100% across 280 tickets; a separate set of 24 hand-written probes scored 20/24. The post includes scripts, manifests, SQL, datasets, and expected outputs in a GitHub repository.
Why it matters
The walkthrough traces captured inference data through Athena analysis, SQL labeling, fine-tuning, and evaluation, including results on held-out and hand-written tickets.
Read next
See the full storyAWS Demonstrates a SageMaker HyperPod Workflow for Capturing Data and Fine-Tuning ModelsThe story so far
AWS demonstrated a SageMaker HyperPod Data Capture workflow that records inference traffic, analyzes it with Athena, applies a SQL labeling policy, and fine-tunes a support router on the same cluster. Data Capture is available on SageMaker HyperPod clusters with the inference operator. In the synthetic walkthrough, a 60-second LoRA fine-tune raised held-out accuracy from 83.9% to 100% across 280 tickets; a separate set of 24 hand-written probes scored 20 out of 24. AWS provided scripts, manifests, SQL, datasets, and expected outputs in a GitHub repository.