Exowatt beats frontier models on accuracy with SageMaker serverless model customization
In this post, we show how Exowatt used Amazon SageMaker AI serverless model customization to build a repeatable AI pipeline that produces task-specific fine-tuned models for two deployment targets, edge and cloud, achieving higher accuracy than frontier models at up to 49x lower cost.
First seen by Oriel on .
AI summary
Exowatt built a repeatable SageMaker AI serverless fine-tuning pipeline whose task-specific Qwen3 models outperformed frontier models while supporting offline edge inference and up to 49× lower costs.
- Fine-tuned Qwen3-1.7B achieved 0.911 macro-F1 on alarm triage, running offline without a GPU as an 8-bit, 1.71 GiB model in a 6 GiB llama.cpp pod.
- Its 2,522-example training set combined 684 verified decisions with 1,838 validated paraphrases; training took about 32 minutes and cost roughly $20–$25.
- Fine-tuned Qwen3-14B reached 0.975 macro-F1 and 1.000 valid JSON on energy-market analysis, versus Claude Opus 4.7 at 0.625 macro-F1.
- The cloud model cost $0.60 per 1,000 answers, compared with $29.26 for Claude Opus 4.7, and runs through Amazon Bedrock Custom Model Import without idle costs.
- The pipeline uses seven AWS serverless services to transform operational data into certified datasets, evaluated models, registry entries, and edge or cloud deployment artifacts within days.
#sagemaker ai #model fine-tuning #edge ai #amazon bedrock
Generated by gpt-5.6-sol. Check the original for details.
Aliases
| Source | External ID | URL |
|---|---|---|
| AWS Machine Learning Blog | 1e1c377670e004fec0787c3710e67b118ba3c87f | https://aws.amazon.com/blogs/machine-learning/exowatt-beats-frontier-models-on-accuracy-with-sagemaker-serverless-model-customization/ Read discussion |