arXiv subjects

Christopher De Sa

Publications and source records attributed to Christopher De Sa.

At least 19 recordsLinked to original sources

arXiv2026

QUASAR: Lowering the Loss Floor of Quantization-Aware Training with Loss-Aware Reconstruction

As large language model inference shifts toward lower precision, post-training quantization (PTQ) becomes increasingly brittle, making quantization-aware training (QAT) essential for preserving model quality. However, QAT computes the loss and surrogate gradients using a lossy reconstruction of latent full-precision weights, while applying updates to the latent weights themselves. This mismatch can lead to suboptimal training trajectories and a higher loss floor. Second-order PTQ methods mitigate a similar gap by minimizing loss-aware reconstruction error, but doing it once for a frozen model can take hours; repeating this process throughout QAT as the weights evolve is impractical. We introduce QUASAR, a QAT method that continuously performs lightweight, loss-aware reconstruction in the training loop to lower the loss floor and improve the resulting low-bit model. At each training step, QUASAR uses the exponential moving average of squared gradients as online saliency estimates, searches over a small set of clipping ranges, and fits affine dequantizers via saliency-weighted least squares. Our analysis shows that the loss-aware reconstruction error is the only reconstruction-dependent term in the QAT convergence bound and controls the loss of the final quantized model, establishing QUASAR's objective as a principled optimization target. QUASAR modifies only the training procedure and supports standard deployment formats, including integer quantization and NVFP4, with no inference-time changes or overhead. Across Qwen3 and Llama-3.1, QUASAR achieves the lowest held-out KL divergence among competitive QAT methods at 2, 3, and 4 bits, reducing KL by at least 10% at 3 and 4 bits and by 29% at 2 bits. At 2 bits, it improves average accuracy across eight tasks by 3.5-4.3 percentage points over strong QAT and PTQ baselines.

cs.LG↗

arXiv2026

Extractable Memorization From First Principles

Recent work on extractable memorization in LLMs suffers from two contrasting validity problems. Some studies overstate extraction, e.g., relying on sequences too short to distinguish memorization from predictability. Others imply that extraction is unreliable evidence of memorization, since models can also reproduce real-world text they weren't explicitly trained on. In different ways, both overlook what makes a valid extraction claim: the model must generate a training sequence with high enough probability to indicate memorization. To determine what's high enough, one has to perform a matched comparison: measuring the generation probabilities of both the training sequences of interest and comparable non-training sequences. Because non-training sequences cannot have been memorized, their probabilities provide a baseline for predictability; a training sequence exceeding this baseline provides evidence of memorization. We formalize matched comparisons in two ways: (1) a conformal test that calibrates a threshold to a chosen FPR when training and non-training sequences are sampled from populations, and (2) a census that calibrates against a matched non-training document when the object is a single document (e.g., a book). We show that matched comparisons enable rigorous, calibrated memorization claims, and reveal where prior setups have validity issues. For instance, on Wikipedia OLMo 2 32B reproduces non-training 10-token suffixes roughly 24% as often as training ones: that share of the training generation rate reflects false positives, not memorization. For Llama 3.1 70B on books, the thresholds we calibrate are as low as 1e-27, supporting memorization claims for sequences that no feasible sampling budget would extract. Based on these results, we refine "extractable memorization" to require a valid memorization claim and near-certain generation within a realistic budget.

cs.LG↗

arXiv2026

Estimating near-verbatim extraction risk in language models with decoding-constrained beam search

Recent work shows that standard greedy-decoding extraction methods for quantifying memorization in LLMs miss how extraction risk varies across sequences. Probabilistic extraction -- computing the probability of generating a target suffix given a prefix under a decoding scheme -- addresses this, but is tractable only for verbatim memorization, missing near-verbatim instances that pose similar privacy and copyright risks. Quantifying near-verbatim extraction risk is expensive: the set of near-verbatim suffixes is combinatorially large, and reliable Monte Carlo (MC) estimation can require ~100,000 samples per sequence. To mitigate this cost, we introduce decoding-constrained beam search, which yields deterministic lower bounds on near-verbatim extraction risk at a cost comparable to ~20 MC samples per sequence. Across experiments, our approach surfaces information invisible to verbatim methods: many more extractable sequences, substantially larger per-sequence extraction mass, and patterns in how near-verbatim extraction risk manifests across model sizes and types of text.

cs.CL↗

arXiv2026

L$^3$: Large Lookup Layers

Modern sparse language models typically achieve sparsity through Mixture-of-Experts (MoE) layers, which dynamically route tokens to dense MLP "experts." However, dynamic hard routing has a number of drawbacks, such as potentially poor hardware efficiency and needing auxiliary losses for stable training. In contrast, the tokenizer embedding table, which is natively sparse, largely avoids these issues by selecting a single embedding per token at the cost of not having contextual information. In this work, we introduce the Large Lookup Layer (L$^3$), which generalizes embedding tables to model decoder layers as a means of further scaling sparsity. L$^3$ layers use static token-based routing to aggregate a set of learned embeddings per token in a context-dependent way, allowing the model to efficiently balance memory and compute by caching information in embeddings. L$^3$ has two main components: (1) a systems-friendly architecture that allows for fast training and CPU-offloaded inference with no overhead, and (2) an information-theoretic embedding allocation algorithm that effectively balances speed and quality. We empirically test L$^3$ by training transformers with up to 2.6B active parameters and find that L$^3$ strongly outperforms both dense models and iso-sparse MoEs in both language modeling and downstream tasks.

cs.LG↗

arXiv2025

Gradient Descent on Logistic Regression: Do Large Step-Sizes Work with Data on the Sphere?

Gradient descent (GD) on logistic regression has many fascinating properties. When the dataset is linearly separable, it is known that the iterates converge in direction to the maximum-margin separator regardless of how large the step size is. In the non-separable case, however, it has been shown that GD can exhibit a cycling behaviour even when the step sizes is still below the stability threshold $2/\lambda$, where $\lambda$ is the largest eigenvalue of the Hessian at the solution. This short paper explores whether restricting the data to have equal magnitude is a sufficient condition for global convergence, under any step size below the stability threshold. We prove that this is true in a one dimensional space, but in higher dimensions cycling behaviour can still occur. We hope to inspire further studies on quantifying how common these cycles are in realistic datasets, as well as finding sufficient conditions to guarantee global convergence with large step sizes.

cs.LG↗

arXiv2025

Model-Preserving Adaptive Rounding

The goal of quantization is to produce a compressed model whose output distribution is as close to the original model's as possible. To do this tractably, most quantization algorithms minimize the immediate activation error of each layer as a proxy for the end-to-end error. However, this ignores the effect of future layers, making it a poor proxy. In this work, we introduce Yet Another Quantization Algorithm (YAQA), an adaptive rounding algorithm that directly considers the error at the network's output. YAQA introduces a series of theoretical results that culminate in the first end-to-end error bounds for quantization algorithms. First, we characterize the convergence time of adaptive rounding algorithms via the structure of their Hessian approximations. We then show that the end-to-end error can be bounded by the approximation's cosine similarity to the true Hessian. This admits a natural Kronecker-factored approximation with corresponding near-optimal Hessian sketches. YAQA is provably better than GPTQ/LDLQ and empirically reduces the error by $\approx 30\%$ over these methods. YAQA even achieves a lower error than quantization aware training. This translates to state of the art performance on downstream tasks, all while adding no inference overhead.

cs.LG↗

arXiv2025

Extracting memorized pieces of (copyrighted) books from open-weight language models

Plaintiffs and defendants in copyright lawsuits over generative AI often make sweeping, opposing claims about the extent to which large language models (LLMs) memorize protected expression from books in their training data. We show that these polarized positions dramatically oversimplify the relationship between memorization and copyright. To do so, we develop a technique to measure memorization of books, which we apply to 200 books and 14 open-weight LLMs. Through over 3000 experiments, we show that memorization varies both by model and book. With respect to our specific extraction methodology, we find that most LLMs do not memorize most books -- either in whole or in part; however, there are notable exceptions. For instance, Llama 3.1 70B entirely memorizes some books, like Harry Potter and the Sorcerer's Stone; memorization is so extensive that one can deterministically extract the whole book almost verbatim using the book's first few words as an initial prompt. We discuss why our results have significant implications for copyright cases, though not ones that unambiguously favor either side.

cs.CL↗

arXiv2025

Compute-Optimal LLMs Provably Generalize Better With Scale

Why do larger language models generalize better? To investigate this question, we develop generalization bounds on the pretraining objective of large language models (LLMs) in the compute-optimal regime, as described by the Chinchilla scaling laws. We introduce a novel, fully empirical Freedman-type martingale concentration inequality that tightens existing bounds by accounting for the variance of the loss function. This generalization bound can be decomposed into three interpretable components: the number of parameters per token, the loss variance, and the quantization error at a fixed bitrate. As compute-optimal language models are scaled up, the number of parameters per data point remains constant; however, both the loss variance and the quantization error decrease, implying that larger models should have smaller generalization gaps. We examine why larger models tend to be more quantizable from an information theoretic perspective, showing that the rate at which they can integrate new information grows more slowly than their capacity on the compute-optimal frontier. From these findings we produce a scaling law for the generalization gap, with bounds that become predictably stronger with scale.

cs.LG↗

arXiv2024

Machine Unlearning Doesn't Do What You Think: Lessons for Generative AI Policy and Research

"Machine unlearning" is a popular proposed solution for mitigating the existence of content in an AI model that is problematic for legal or moral reasons, including privacy, copyright, safety, and more. For example, unlearning is often invoked as a solution for removing the effects of specific information from a generative-AI model's parameters, e.g., a particular individual's personal data or the inclusion of copyrighted content in the model's training data. Unlearning is also proposed as a way to prevent a model from generating targeted types of information in its outputs, e.g., generations that closely resemble a particular individual's data or reflect the concept of "Spiderman." Both of these goals--the targeted removal of information from a model and the targeted suppression of information from a model's outputs--present various technical and substantive challenges. We provide a framework for ML researchers and policymakers to think rigorously about these challenges, identifying several mismatches between the goals of unlearning and feasible implementations. These mismatches explain why unlearning is not a general-purpose solution for circumscribing generative-AI model behavior in service of broader positive impact.

cs.LG↗

arXiv2024

Searching for Efficient Linear Layers over a Continuous Space of Structured Matrices

Dense linear layers are the dominant computational bottleneck in large neural networks, presenting a critical need for more efficient alternatives. Previous efforts focused on a small number of hand-crafted structured matrices and neglected to investigate whether these structures can surpass dense layers in terms of compute-optimal scaling laws when both the model size and training examples are optimally allocated. In this work, we present a unifying framework that enables searching among all linear operators expressible via an Einstein summation. This framework encompasses many previously proposed structures, such as low-rank, Kronecker, Tensor-Train, Block Tensor-Train (BTT), and Monarch, along with many novel structures. To analyze the framework, we develop a taxonomy of all such operators based on their computational and algebraic properties and show that differences in the compute-optimal scaling laws are mostly governed by a small number of variables that we introduce. Namely, a small $\omega$ (which measures parameter sharing) and large $\psi$ (which measures the rank) reliably led to better scaling laws. Guided by the insight that full-rank structures that maximize parameters per unit of compute perform the best, we propose BTT-MoE, a novel Mixture-of-Experts (MoE) architecture obtained by sparsifying computation in the BTT structure. In contrast to the standard sparse MoE for each entire feed-forward network, BTT-MoE learns an MoE in every single linear layer of the model, including the projection matrices in the attention blocks. We find BTT-MoE provides a substantial compute-efficiency gain over dense layers and standard MoE.

cs.LG↗

arXiv2024

Non-Determinism and the Lawlessness of Machine Learning Code

Legal literature on machine learning (ML) tends to focus on harms, and thus tends to reason about individual model outcomes and summary error rates. This focus has masked important aspects of ML that are rooted in its reliance on randomness -- namely, stochasticity and non-determinism. While some recent work has begun to reason about the relationship between stochasticity and arbitrariness in legal contexts, the role of non-determinism more broadly remains unexamined. In this paper, we clarify the overlap and differences between these two concepts, and show that the effects of non-determinism, and consequently its implications for the law, become clearer from the perspective of reasoning about ML outputs as distributions over possible outcomes. This distributional viewpoint accounts for randomness by emphasizing the possible outcomes of ML. Importantly, this type of reasoning is not exclusive with current legal reasoning; it complements (and in fact can strengthen) analyses concerning individual, concrete outcomes for specific automated decisions. By illuminating the important role of non-determinism, we demonstrate that ML code falls outside of the cyberlaw frame of treating ``code as law,'' as this frame assumes that code is deterministic. We conclude with a brief discussion of what work ML can do to constrain the potentially harm-inducing effects of non-determinism, and we indicate where the law must do work to bridge the gap between its current individual-outcome focus and the distributional approach that we recommend.

cs.CY↗

arXiv2024

QTIP: Quantization with Trellises and Incoherence Processing

Post-training quantization (PTQ) reduces the memory footprint of LLMs by quantizing weights to low-precision datatypes. Since LLM inference is usually memory-bound, PTQ methods can improve inference throughput. Recent state-of-the-art PTQ approaches use vector quantization (VQ) to quantize multiple weights at once, which improves information utilization through better shaping. However, VQ requires a codebook with size exponential in the dimension. This limits current VQ-based PTQ works to low VQ dimensions ($\le 8$) that in turn limit quantization quality. Here, we introduce QTIP, which instead uses trellis coded quantization (TCQ) to achieve ultra-high-dimensional quantization. TCQ uses a stateful decoder that separates the codebook size from the bitrate and effective dimension. QTIP introduces a spectrum of lookup-only to computed lookup-free trellis codes designed for a hardware-efficient "bitshift" trellis structure; these codes achieve state-of-the-art results in both quantization quality and inference speed.

cs.LG↗

arXiv2024

Gradient Descent on Logistic Regression with Non-Separable Data and Large Step Sizes

We study gradient descent (GD) dynamics on logistic regression problems with large, constant step sizes. For linearly-separable data, it is known that GD converges to the minimizer with arbitrarily large step sizes, a property which no longer holds when the problem is not separable. In fact, the behaviour can be much more complex -- a sequence of period-doubling bifurcations begins at the critical step size $2/\lambda$, where $\lambda$ is the largest eigenvalue of the Hessian at the solution. Using a smaller-than-critical step size guarantees convergence if initialized nearby the solution: but does this suffice globally? In one dimension, we show that a step size less than $1/\lambda$ suffices for global convergence. However, for all step sizes between $1/\lambda$ and the critical step size $2/\lambda$, one can construct a dataset such that GD converges to a stable cycle. In higher dimensions, this is actually possible even for step sizes less than $1/\lambda$. Our results show that although local convergence is guaranteed for all step sizes less than the critical step size, global convergence is not, and GD may instead converge to a cycle depending on the initialization.

cs.LG↗

arXiv2024

Zeroth-Order Fine-Tuning of LLMs with Extreme Sparsity

Zeroth-order optimization (ZO) is a memory-efficient strategy for fine-tuning Large Language Models using only forward passes. However, the application of ZO fine-tuning in memory-constrained settings such as mobile phones and laptops is still challenging since full precision forward passes are infeasible. In this study, we address this limitation by integrating sparsity and quantization into ZO fine-tuning of LLMs. Specifically, we investigate the feasibility of fine-tuning an extremely small subset of LLM parameters using ZO. This approach allows the majority of un-tuned parameters to be quantized to accommodate the constraint of limited device memory. Our findings reveal that the pre-training process can identify a set of "sensitive parameters" that can guide the ZO fine-tuning of LLMs on downstream tasks. Our results demonstrate that fine-tuning 0.1% sensitive parameters in the LLM with ZO can outperform the full ZO fine-tuning performance, while offering wall-clock time speedup. Additionally, we show that ZO fine-tuning targeting these 0.1% sensitive parameters, combined with 4 bit quantization, enables efficient ZO fine-tuning of an Llama2-7B model on a GPU device with less than 8 GiB of memory and notably reduced latency.

cs.LG↗

arXiv2024

QuIP#: Even Better LLM Quantization with Hadamard Incoherence and Lattice Codebooks

Post-training quantization (PTQ) reduces the memory footprint of LLMs by quantizing their weights to low-precision. In this work, we introduce QuIP#, a weight-only PTQ method that achieves state-of-the-art results in extreme compression regimes ($\le$ 4 bits per weight) using three novel techniques. First, QuIP# improves QuIP's (Chee et al., 2023) incoherence processing by using the randomized Hadamard transform, which is faster and has better theoretical properties. Second, QuIP# uses vector quantization to take advantage of the ball-shaped sub-Gaussian distribution that incoherent weights possess: specifically, we introduce a set of hardware-efficient codebooks based on the highly symmetric $E_8$ lattice, which achieves the optimal 8-dimension unit ball packing. Third, QuIP# uses fine-tuning to improve fidelity to the original model. Our experiments show that QuIP# outperforms existing PTQ methods, enables new behaviors in PTQ scaling, and supports fast inference. Our code can be found at https://github.com/Cornell-RelaxML/quip-sharp.

cs.LG↗

arXiv2024

STAT: Shrinking Transformers After Training

We present STAT: a simple algorithm to prune transformer models without any fine-tuning. STAT eliminates both attention heads and neurons from the network, while preserving accuracy by calculating a correction to the weights of the next layer. Each layer block in the network is compressed using a series of principled matrix factorizations that preserve the network structure. Our entire algorithm takes minutes to compress BERT, and less than three hours to compress models with 7B parameters using a single GPU. Using only several hundred data examples, STAT preserves the output of the network and improves upon existing gradient-free pruning methods. It is even competitive with methods that include significant fine-tuning. We demonstrate our method on both encoder and decoder architectures, including BERT, DistilBERT, and Llama-2 using benchmarks such as GLUE, Squad, WikiText2.

cs.LG↗

arXiv2024

Shadow Cones: A Generalized Framework for Partial Order Embeddings

Hyperbolic space has proven to be well-suited for capturing hierarchical relations in data, such as trees and directed acyclic graphs. Prior work introduced the concept of entailment cones, which uses partial orders defined by nested cones in the Poincaré ball to model hierarchies. Here, we introduce the ``shadow cones" framework, a physics-inspired entailment cone construction. Specifically, we model partial orders as subset relations between shadows formed by a light source and opaque objects in hyperbolic space. The shadow cones framework generalizes entailment cones to a broad class of formulations and hyperbolic space models beyond the Poincaré ball. This results in clear advantages over existing constructions: for example, shadow cones possess better optimization properties over constructions limited to the Poincaré ball. Our experiments on datasets of various sizes and hierarchical structures show that shadow cones consistently and significantly outperform existing entailment cone constructions. These results indicate that shadow cones are an effective way to model partial orders in hyperbolic space, offering physically intuitive and novel insights about the nature of such structures.

cs.LG↗

arXiv2024

ModuLoRA: Finetuning 2-Bit LLMs on Consumer GPUs by Integrating with Modular Quantizers

We propose a memory-efficient finetuning algorithm for large language models (LLMs) that supports finetuning LLMs with 65B parameters in 2/3/4-bit precision on as little as one 24GB GPU. Our method, modular low-rank adaptation (ModuLoRA), integrates any user-specified weight quantizer with finetuning via low-rank adapters (LoRAs). Our approach relies on a simple quantization-agnostic backward pass that adaptively materializes low-precision LLM weights from a custom black-box quantization module. This approach enables finetuning 2-bit and 3-bit LLMs for the first time -- leveraging state-of-the-art 2-bit QuIP\# quantization and 3-bit OPTQ quantization -- outperforming finetuning that relies on less sophisticated 4-bit and 8-bit methods. In our experiments, \lplora~attains competitive performance on text classification, natural language inference, and instruction following tasks using significantly less memory than existing approaches, and we also surpass the state-of-the-art ROUGE score on a popular summarization task. We release \lplora~together with a series of low-precision models as part of \llmtune, a user-friendly library for quantizing, running, and finetuning LLMs on consumer GPUs.

cs.LG↗