Write your own training loop — without managing GPUs — with Amazon SageMaker training session APIs

Key points

• Sometimes a standard recipe isn’t enough: you need a custom loss or data curriculum for supervised fine-tuning (SFT), your own preference optimization, a reinforcement learning (RL) reward function, or an alignment algorithm published last week.1
• Today, serverless model customization in Amazon SageMaker AI expands with training session APIs, which remove that tradeoff by drawing the abstraction boundary one level lower.1
• Operations like forwardbackward(), optimstep(), and sample() become SageMaker API calls that you compose into whatever procedure your technique requires: SFT, preference optimization, RL, or something new.1
• In this post, we introduce the programming model and the three SDK clients, and show where training session APIs fit among the model training and customization options in SageMaker AI.1

Sources (1)

• [1]Write your own training loop — without managing GPUs — with Amazon SageMaker training session APIs
  AWS Machine Learning Blog · Oct 22, 06:00 PM

  Fine-tuning a large language model (LLM) today usually forces a tradeoff between training loop control and operational simplicity.

  Sometimes a standard recipe isn’t enough: you need a custom loss or data curriculum for supervised fine-tuning (SFT), your own preference optimization, a reinforcement learning (RL) reward function, or an alignment algorithm published last week.

Extractive summary: sentences quoted from the sources.