Isaac Rehg

Anh Thai, Stefan Stojanov, Zixuan Huang et al.

Continual learning has been extensively studied for classification tasks with methods developed to primarily avoid catastrophic forgetting, a phenomenon where earlier learned concepts are forgotten at the expense of more recent samples. In this work, we present a set of continual 3D object shape reconstruction tasks, including complete 3D shape reconstruction from different input modalities, as well as visible surface (2.5D) reconstruction which, surprisingly demonstrate positive knowledge (backward and forward) transfer when training with solely standard SGD and without additional heuristics. We provide evidence that continuously updated representation learning of single-view 3D shape reconstruction improves the performance on learned and novel categories over time. We provide a novel analysis of knowledge transfer ability by looking at the output distribution shift across sequential learning tasks. Finally, we show that the robustness of these tasks leads to the potential of having a proxy representation learning task for continual classification. The codebase, dataset and pre-trained models released with this article can be found at https://github.com/rehg-lab/CLRec

Isaac Rehg

Context lengths of Large Language Models (LLMs) have exploded in recent years, with 128k-token context becoming a standard and million-token context becoming a reality. Efficiently supporting long-context inference remains challenging as the memory that must be allocated in key-value (KV) cache for a generation scales with its context length, limiting the number of long-context requests that can be served concurrently under a given memory budget. KV cache compression can mitigate this issue by removing under-utilized KVs from each attention head's cache and reducing its memory footprint. Higher theoretical compression rates can be achieved when the number of removed KVs varies across attention heads, but application of such a strategy within existing inference frameworks adds fragmentation and cannot realize the theoretical compression rates in physical memory. We introduce KV-Compress, a novel compression method that evicts contiguous KV blocks within a PagedAttention framework, reducing the memory footprint of the KV cache proportionally to this theoretical compression rate. Our method achieves state-of-the-art performance on LongBench for both Mistral-7B-Instruct-v0.2 and Llama-3.1-8B-Instruct while lowering the total number of compressed KVs by 4x compared with prior methods. Evaluations on Llama-3.1-8B-Instruct and Llama-3.1-70B-Instruct-FP8 achieve compression rates up to 8x with negligible impact on performance, and up to 64x while retaining over 90% of full-cache performance for all but three of the suite's subsets. We benchmark an integration of our method with vLLM that increases total throughput by up to 5.18x by enabling larger decoding batches.

Yinghao Li, Rui Feng, Isaac Rehg et al.

We study the problem of using (partial) constituency parse trees as syntactic guidance for controlled text generation. Existing approaches to this problem use recurrent structures, which not only suffer from the long-term dependency problem but also falls short in modeling the tree structure of the syntactic guidance. We propose to leverage the parallelism of Transformer to better incorporate parse trees. Our method first expands a partial template constituency parse tree to a full-fledged parse tree tailored for the input source text, and then uses the expanded tree to guide text generation. The effectiveness of our model in this process hinges upon two new attention mechanisms: 1) a path attention mechanism that forces one node to attend to only other nodes located in its path in the syntax tree to better incorporate syntax guidance; 2) a multi-encoder attention mechanism that allows the decoder to dynamically attend to information from multiple encoders. Our experiments in the controlled paraphrasing task show that our method outperforms SOTA models both semantically and syntactically, improving the best baseline's BLEU score from 11.83 to 26.27.