# PROOF-Gen: From Optimized Data to Better Distillation

DevFeed: [PROOF-Gen: From Optimized Data to Better Distillation](<https://devfeed.tech/articles/proof-gen-from-optimized-data-to-better-distillation-6731.md>)

Original publisher: [Read original article](<https://machinelearning.apple.com/research/proof-gen-optimized-distillation>)

Published: 2026-08-26T00:00:00Z

Content type: article

Language: en

Sources: [Apple Machine Learning Research](<https://devfeed.tech/sources/apple-machine-learning-research.md>)

Topics: [Fine-tuning](<https://devfeed.tech/topics/fine-tuning.md>), [Prompt optimization](<https://devfeed.tech/topics/prompt-optimization.md>), [Artificial Intelligence](<https://devfeed.tech/topics/ai.md>), [gemma4](<https://devfeed.tech/topics/gemma4.md>), [Programming](<https://devfeed.tech/topics/programming.md>)

Tags: [agent](<https://devfeed.tech/tags/agent.md>), [ai-agents](<https://devfeed.tech/tags/ai-agents.md>), [evaluation](<https://devfeed.tech/tags/evaluation.md>), [fine-tuning](<https://devfeed.tech/tags/fine-tuning.md>), [gemma](<https://devfeed.tech/tags/gemma.md>), [generate](<https://devfeed.tech/tags/generate.md>), [models](<https://devfeed.tech/tags/models.md>), [on-device](<https://devfeed.tech/tags/on-device.md>), [post-training](<https://devfeed.tech/tags/post-training.md>), [prompt-optimization](<https://devfeed.tech/tags/prompt-optimization.md>)

## AI overview

PROOF-Gen improves tool-calling model distillation by using per-scenario prompt optimization to recover successful trajectories from failed teacher attempts. The method strips corrective guidance before training, producing clean demonstrations and improving benchmark, deployed-pipeline, and on-device model performance.

## Source excerpt

Supervised fine-tuning on teacher-generated trajectories is the standard first stage for distilling tool-calling capabilities into deployable models. Post-training pipelines that drive shipped tool-calling agents re-run this stage on a daily or weekly cadence, paying the frontier-teacher cost each cycle, yet the mechanism is generate-and-filter (keep the teacher's passing trajectories, discard the rest) and each cycle leaves behind the same hard scenarios because failures supply no signal. On τ 2-bench, 57% of teacher trials fail, two-thirds of them near-misses (most tool calls correct, undone...