Michael Helcig

Michael Helcig, Dan Alistarh

Assigning one of K options to each of N groups under a total cost budget is a recurring problem in machine learning, appearing in mixed-precision quantization, non-uniform pruning, and expert selection. The objective (model loss) depends jointly on all assignments and does not decompose across groups, which prevents combinatorial solvers from optimizing the true objective directly and limits them to proxy objectives. Evolutionary search evaluates the actual loss but lacks gradient information, while penalty-based methods provide gradients but enforce the budget only approximately and require sensitive hyperparameter tuning. We observe that under softmax relaxation, the budget constraint defines a smooth Riemannian manifold in logit space with particularly simple geometry: the normal vector is available in closed form, shifting logits along the cost vector changes expected cost monotonically, allowing binary-search retraction, and vector transport reduces to a single inner product. Building on this structure, we propose Riemannian Constrained Optimization (RCO), which augments a standard Adam update with tangent projection, binary-search retraction, and momentum transport. Combined with Gumbel straight-through estimation and budget-constrained dynamic programming for discrete feasibility, RCO enables first-order optimization of the true objective under exact budget enforcement, without introducing constraint hyperparameters. On synthetic knapsack problems with known optima, the manifold-based constraint handling recovers optimal solutions, whereas penalty methods plateau at 83% of optimal. On LLM compression tasks, including mixed-precision quantization and MoE expert pruning, RCO matches or exceeds evolutionary search methods while requiring 3x to 16x lower wall-clock cost on the evaluated configurations.

Alireza Dadgarnia, Soroush Tabesh, Mahdi Nikdan et al.

Weight quantization has become a standard tool for efficient LLM deployment, especially for local inference, where models are now routinely served at 2-3 bits per parameter. The state of the art is currently split into two sets of methods: simple scalar quantization techniques, such as GPTQ or AWQ, which are widely deployed but plateau in accuracy at 3-4 bits per parameter (bpp), and "second-generation" vector- or trellis-quantized methods, such as QTIP, GPTVQ and AQLM, which push the accuracy frontier at low bit-widths but are notoriously hard to implement and to scale, and have gained relatively less traction. In this paper, we ask whether this gap is fundamental, or whether a carefully optimized scalar quantizer can recover most of it. We answer in the affirmative, by introducing GSQ (Gumbel-Softmax Quantization), a post-training scalar quantization method which jointly learns the per-coordinate grid assignments and the per-group scales using a Gumbel-Softmax relaxation of the discrete grid. GSQ matches the cardinality of the relaxation to the small number of levels available in the target bit-width regime (e.g., 3-8 levels for ternary and 3 bpp, respectively), making the relaxation tight and the optimization tractable. Practically, on the standard Llama-3.1-8B/70B-Instruct models, GSQ closes most of the gap between scalar quantization and the QTIP frontier at 2 and 3 bits, while using a symmetric scalar grid with group-wise quantization, and thus fully compatible with existing scalar inference kernels. We further show that GSQ scales to trillion-scale Mixture-of-Experts models such as Kimi-K2.5, where vector-quantized methods are difficult to apply.

Michael Helcig, Eldar Kurtic, Dan Alistarh

Model quantization has become essential for efficient large language model deployment, yet existing approaches involve clear trade-offs: methods such as GPTQ and AWQ achieve practical compression but are lossy, while lossless techniques preserve fidelity but typically do not accelerate inference. This paper explores the middle ground of statistically-lossless compression through three complementary notions of losslessness for quantized LLMs. First, task-lossless compression preserves zero-shot benchmark accuracy within natural sampling variance and remains achievable at aggressive bitwidths. Second, we formalize the stricter notion of distribution-lossless compression, requiring the quantized model's next-token distribution to be practically indistinguishable from the original, and propose the Expected Acceptance Rate (EAR), the maximum token-agreement probability under optimal coupling, as a directly interpretable fidelity metric (for example, EAR >= 0.99 indicates 99% agreement). Third, we prove a gamma-squared variance law showing that symmetric quantization inflates noise variance by gamma squared relative to asymmetric quantization, making asymmetry necessary for distribution-lossless fidelity but not for task-level preservation. Using SLQ, a layer-wise non-uniform method with asymmetric quantization and wide bitwidth search, we achieve task-lossless compression at well below 4 bits per parameter (as low as 3.3 bits depending on the model), distribution-lossless compression at 5 to 6 bits per parameter on average, and inference speedups of 1.7 to 3.6x relative to FP16 with optimized kernels. Source code is available at https://github.com/IST-DASLab/SLQ.