Impossible Distillation: from Low-Quality Model to High-Quality Dataset & Model for Summarization and Paraphrasing
This work addresses the challenge of resource-efficient text generation for NLP practitioners by enabling high-quality outputs from smaller models, though it is incremental in leveraging existing pre-trained LMs.
The paper tackles the problem of generating high-quality datasets and models for paraphrasing and summarization from low-quality teacher models that cannot perform these tasks, achieving results where a 770M parameter model outperforms strong baselines including models distilled from ChatGPT and sometimes ChatGPT itself, with a distilled dataset from 1.5B LMs showing higher diversity and fidelity than datasets up to 13 times larger.
We present Impossible Distillation, a novel framework for paraphrasing and sentence summarization, that distills a high-quality dataset and model from a low-quality teacher that itself cannot perform these tasks. Unlike prior works that rely on an extreme-scale teacher model (e.g., GPT3) or task-specific architecture, we hypothesize and verify the paraphrastic proximity intrinsic to pre-trained LMs (e.g., GPT2), where paraphrases occupy a proximal subspace in the LM distribution. By identifying and distilling generations from these subspaces, Impossible Distillation produces a high-quality dataset and model even from GPT2-scale LMs. We evaluate our method on multiple benchmarks spanning unconstrained / syntax-controlled paraphrase generation and sentence summarization. Our model with 770M parameters consistently outperforms strong baselines, including models distilled from ChatGPT, and sometimes, even ChatGPT itself. Also, we find that our distilled dataset from 1.5B LMs exhibits higher diversity and fidelity than up to 13 times larger datasets.