LLM-to-Flow Distillation: Teaching Natural Language Goals to Scientific Generators

Key points

• Yet these capabilities remain largely separated: foundation LLMs encode broad chemical and biological knowledge, while domain-specific flow- and diffusion-based generators capture the priors needed to produce valid sequences and geometries.1
• We study how to align such scientific generators to high-level natural-language discovery goals without requiring an explicit proxy reward, which is often unavailable or poorly captured by learned predictors in experimental sciences.1
• We introduce LLM-to-Flow Distillation (LFD), where a frozen LLM acts as an adaptive teacher for a domain-specific generator.1
• Experiments on therapeutic peptides and small-molecule structure design demonstrate that LFD enables a new form of generative control: aligning arbitrary domain-specific scientific generators to scientific discovery tasks expressed via high-level natural-language goals.1

Sources (2)

• [1]LLM-to-Flow Distillation: Teaching Natural Language Goals to Scientific Generators
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 12, 06:11 PM

  Generative discovery in science requires coupling high-level scientific reasoning with the ability to generate samples respecting the structure of a design space.

  Yet these capabilities remain largely separated: foundation LLMs encode broad chemical and biological knowledge, while domain-specific flow- and diffusion-based generators capture the priors needed to produce valid sequences and geometries.

• [2]LLM-to-Flow Distillation: Teaching Natural Language Goals to Scientific Generators
  Hugging Face Daily Papers · Oct 12, 12:00 AM · same content

Extractive summary: sentences quoted from the sources.