Llama 4 Scout

Microsoft Azure AI Foundry introduces new fine-tuning options, including Supervised Fine-Tuning (SFT) for GPT-4.1-nano and Llama 4 Scout

RFT uses a feedback loop and custom grading functions to align model behavior with complex business logic and domain-specific requirements

RFT enables models to learn not just what to output, but why, improving decision-making in dynamic or high-stakes scenarios

The o4-mini model is the first compact reasoning model in Azure AI Foundry to support RFT, excelling at structured reasoning and chain-of-thought prompt

RFT-tuned models offer improved contextual awareness, adaptive reasoning, and require fewer training examples to match supervised approaches

RFT is ideal for domains with custom rules, operational standards, or high decision-making complexity, such as legal, healthcare, and finance

Early adopters (e.g., DraftWise, Accordance AI, Ambience Healthcare, Harvey, Runloop, Milo, SafetyKit) have reported significant improvements in accuracy and output quality

RFT setup involves designing a Python grading function, preparing a prompt dataset, launching training via API or dashboard, and iterating based on results

RFT with o4-mini will be available in Sweden Central and East US2, with pricing at $100 per hour of core training time and a 50% rebate for research dataset contributions

SFT for GPT-4.1-nano allows organizations to tailor the model to their domain, supporting high-throughput, low-latency, and cost-sensitive workloads.

For more details visit Govindhtech.com