AI Daybook
AI Daybook
Today's issue
PaperResearch

PLatour: evaluating collaborative prompt-generation transfer in language models

Original titleOn the Opportunities and Challenges of Prefill-to-Autoregressive Cross-Model Transfer

Outlet
Hugging Face
Published
Importance
2/5
Trending
10

Summary

The authors propose PLatour, an open-source framework for Prefill-to-Autoregressive Cross-model Transfer (PACT), which combines a larger model's prompt representations with a smaller model's efficient generation to balance capability and inference cost. PLatour connects frozen models through lightweight linear bridges trained with task rewards and supports hybrid-attention architectures. It significantly outperforms open-source baselines on generation tasks and surpasses prior results on shared mathematics and coding benchmarks, though controlled experiments show some gains can be reproduced by uninformative sender signals or receiver-only training.

Original excerpt

Prefill-to-Autoregressive Cross-model Transfer (PACT), a form of latent communication between language models, combines a larger model's prompt representations with a smaller model's efficient generation, offering a promising way to balance capability and inference cost. However, existing systems provide limited support for independently evaluating large-small model collaboration, and improved task performance alone does not establish whether the receiver benefits from the sender's information. We introduce PLatour, an open-source PACT framework that connects frozen models through lightweight, zero-initialized linear bridges trained directly with task rewards and supports hybrid-attention architectures. PLatour significantly outperforms open-source baselines on generation tasks and surpasses Mostik's reported results on shared mathematics and coding benchmarks using a substantially smaller sender. However, controlled experiments show that the gains on some tasks can be reproduced by uninformative sender signals, self-communication, or receiver-only training. We find that trainable bridges can act as parameter-efficient adaptation mechanisms without exploiting useful sender information. We therefore propose PACT-Check to quantify how much of a trained PACT system's gain genuinely comes from communication. Finally, we discuss the opportunities and challenges of PACT systems. Code is available at https://github.com/zhenyi4/platour.

#llm#inference#efficiency#transfer-learning
Read the original
See all →