Apple unveils LenVM for fine-grained token-level length modeling in LLMs

Apple researchers have introduced the Length Value Model (LenVM), a novel framework for token-level length modeling in autoregressive models. This approach treats length prediction as a value estimation problem, assigning a negative reward to each token to create a scalable and annotation-free supervision signal. Experiments show LenVM significantly improves length matching performance on benchmarks like LIFEBench and enhances efficiency on tasks such as GSM8K, allowing for controlled trade-offs between accuracy and token budget. AI

IMPACT Introduces a new method for controlling generation length and efficiency in LLMs, potentially impacting inference costs and model performance.

RANK_REASON Academic paper detailing a new modeling technique for LLMs. [lever_c_demoted from research: ic=1 ai=1.0]

Read on Apple Machine Learning Research →

• Apple Inc.
• Carnegie Mellon University
• GSM8K
• Length Value Model
• LenVM
• LIFEBench
• University of California, Santa Barbara
• University of Wisconsin–Madison

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →