Entity · paper

How Much is Left? LLMs Linearly Encode Their Remaining Output Length

paperactiveprovisionalhow-much-is-left-llms-linearly-encode-their-remaining-output-length-236b420e·1 events·first seen 41h ago

Aliases: How Much is Left? LLMs Linearly Encode Their Remaining Output Length

More like this (12)

Continual LLM Upcycling: A Predictor-Gated Bank-Wise Sparsity Training Recipe for Dense-to-Sparse LLMs code synthesis LLMs A sleep-like consolidation mechanism for LLMs Feature-wise Linear Modulation CLP: Collocation-Length Prediction for Zero-Loss Adaptive Multi-Token Inference Knowledge Knows, Verbalization Tells: Disentangling Latent Directions for Mathematical Solvability in LLMs long-context LLMs AIR: Adaptive Interleaved Reasoning with Code in MLLMs Follow the Latent Roadmap: Navigating Revocable Decoding for Diffusion LLMs with Anchor Tokens LLM inference llms.txt CLP (Collocation-Length Predictor)

Recent events (1)

6arXiv · cs.CL·41h ago·source ↗

Linear probes reveal LLMs encode remaining output length in hidden states

Researchers trained minimal linear probes on frozen hidden states of three open-weight 7-8B models and found that total response length is linearly decodable from the prompt's final hidden state before any output is generated. The probe directions transfer across natural-language and synthetic datasets, and per-position estimates shift upward when models retract and restart partial solutions. The authors interpret this as evidence that LLMs maintain a plan-like internal representation of remaining generation length, distinct from exact-counting, though causality is not established.

Evaluation and Benchmarking How Much is Left? LLMs Linearly Encode Their Remaining Output Length