arXiv subjects

Allan Kazakov

Publications and source records attributed to Allan Kazakov.

4 recordsLinked to original sources

arXiv2026

Accepted Prefixes Are Not All You Need: A Negative Result on PEFT-Based Block-Diffusion Drafting

Speculative decoding accelerates autoregressive language model inference by using a cheap drafter to propose multiple future tokens and a target model to verify them. A common design goal is therefore to improve draft quality while reducing auxiliary parameters and systems overhead. We study a negative result for this direction through PEFT-BD, a same-backbone speculative decoding method in which a LoRA-like adapter acts as a block-diffusion drafter for an autoregressive verifier. PEFT-BD is motivated by several attractive properties: it avoids tokenizer mismatch, avoids loading a separate draft model, adds only a small number of trainable parameters, and uses a BD3LM-style denoising objective to propose a block of tokens in parallel. Despite these advantages, PEFT-BD does not yield a practical speedup in our Qwen3-0.6B experiments. Although the method obtains nontrivial accepted prefixes, profiling shows that each speculative step requires an adapter-enabled full-backbone draft pass followed by an adapter-disabled full-backbone verification pass. Thus, the drafter is parameter-efficient but not compute-efficient. Our results isolate a simple but important condition for successful speculative decoding: the drafter must be substantially cheaper to execute than the verifier. Longer accepted prefixes alone cannot compensate when draft computation remains verifier-scale.

cs.AI↗

arXiv2026

Revitalizing Dense Material Segmentation: Stabilized Vision Transformers and the Generalization Paradox

Material segmentation, the pixel-wise classification of physical surface properties, remains a challenging problem in computer vision, requiring physicochemical understanding distinct from object-centric parsing. Despite the introduction of the rigorous Apple Dense Material Segmentation (DMS) dataset, the benchmark has suffered from attrition and stagnation, increasingly overshadowed by geometry-biased foundation models. In this paper, we revive the Apple-DMS benchmark to establish a modern Vision Transformer baseline. We conduct an exhaustive evaluation of SegFormer and Mask2Former architectures, revealing that standard training paradigms fail on amorphous texture fields due to high-variance gradients. To address this, we introduce a stabilized training recipe featuring High-Fidelity Logit Projection, Query Entropy Regularization, and a domain-specific, physics-compliant augmentation pipeline. Our optimized SegFormer-B5 achieves a new State-of-the-Art (SOTA) of 0.4572 mIoU on the original dataset split, significantly surpassing the prior convolutional baseline. Furthermore, we identify a critical "Generalization Paradox": while re-partitioning the dataset into a data-rich 80/10/10 split inflates the metric to 0.5276 mIoU, expert qualitative analysis reveals this induces distributional homogenization, severely degrading real-world, out-of-distribution performance. By releasing our recovered dataset index and robust training framework, we demonstrate that material perception is far from solved and urge the community to leverage the rigorous original split to drive genuine progress in physically grounded artificial intelligence.

cs.CV↗

arXiv2026

Silicon Showdown: Performance, Efficiency, and Ecosystem Barriers in Consumer-Grade LLM Inference

The operational landscape of local Large Language Model (LLM) inference has shifted from lightweight models to datacenter-class weights exceeding 70B parameters, creating profound systems challenges for consumer hardware. This paper presents a systematic empirical analysis of the Nvidia and Apple Silicon ecosystems, specifically characterizing the distinct intra-architecture trade-offs required to deploy these massive models. On the Nvidia Blackwell architecture, we identify a critical "Backend Dichotomy" within the TensorRT-LLM stack: while the new NVFP4 quantization format delivers a 1.6x throughput advantage over optimized BF16 baselines (151 tokens/s vs. 92 tokens/s), realizing this performance requires navigating complex runtime constraints that trade startup latency for generation speed. Furthermore, we characterize the "VRAM Wall" for 70B+ models: on discrete GPUs, users face a destructive choice between aggressive quantization (e.g., Q2) that degrades model intelligence to fit in VRAM, or PCIe-bottlenecked CPU offloading, which reduces throughput by over 90% compared to full-GPU execution. Conversely, Apple's Unified Memory Architecture (UMA) circumvents these bottlenecks, enabling linear scaling for 80B parameter models at practical 4-bit precisions. This architectural divergence extends to operational sustainability, where Apple's SoC design demonstrates up to a 23x advantage in energy efficiency (tokens/joule). We conclude that for consumer-grade inference, the optimal hardware is defined by a complex interplay between compute density (Nvidia) and memory capacity (Apple), moderated by the significant "ecosystem friction" of proprietary quantization workflows.

cs.PF↗

arXiv2025

Quantization Robustness to Input Degradations for Object Detection

Post-training quantization (PTQ) is crucial for deploying efficient object detection models, like YOLO, on resource-constrained devices. However, the impact of reduced precision on model robustness to real-world input degradations such as noise, blur, and compression artifacts is a significant concern. This paper presents a comprehensive empirical study evaluating the robustness of YOLO models (nano to extra-large scales) across multiple precision formats: FP32, FP16 (TensorRT), Dynamic UINT8 (ONNX), and Static INT8 (TensorRT). We introduce and evaluate a degradation-aware calibration strategy for Static INT8 PTQ, where the TensorRT calibration process is exposed to a mix of clean and synthetically degraded images. Models were benchmarked on the COCO dataset under seven distinct degradation conditions (including various types and levels of noise, blur, low contrast, and JPEG compression) and a mixed-degradation scenario. Results indicate that while Static INT8 TensorRT engines offer substantial speedups (~1.5-3.3x) with a moderate accuracy drop (~3-7% mAP50-95) on clean data, the proposed degradation-aware calibration did not yield consistent, broad improvements in robustness over standard clean-data calibration across most models and degradations. A notable exception was observed for larger model scales under specific noise conditions, suggesting model capacity may influence the efficacy of this calibration approach. These findings highlight the challenges in enhancing PTQ robustness and provide insights for deploying quantized detectors in uncontrolled environments. All code and evaluation tables are available at https://github.com/AllanK24/QRID.

cs.CV↗