arXiv subjects

Chengqiu Hu

Publications and source records attributed to Chengqiu Hu.

5 recordsLinked to original sources

arXiv2026

CANN Bench: Benchmarking Agent Generated Kernels against Real NPU and Algorithmic Limits

AI agents are now capable of writing, compiling, and iteratively optimizing low-level operator kernels on different hardware platforms. Existing benchmarks, however, focus almost exclusively on CUDA and Triton, leaving hardware ecosystems with less-exposed programming models without a common evaluation baseline. We present CANN Bench, an open benchmark for AI-generated operator code on Huawei's Ascend NPU. The current release covers 53 operators and 1060 test cases organized into four difficulty tiers -- from simple elementwise primitives to MoE dispatch and FlashAttention kernels -- spanning FP16, BF16, FP32, and INT8 precision formats. Evaluation adopts a \textbf{three-dimensional weighted composite score} that treats compilation, functional correctness, and performance as independent axes, providing a principled reward signal for kernel-generation agents. Performance is graded against an out-of-the-box PyTorch-on-Ascend baseline and an analytical per-case Hardware-Anchored Performance (HAP) limit on real NPU hardware, ensuring scores reflect genuine optimization headroom rather than measurement artifacts. The evaluation harness is designed to resist reward hacking from the ground up. CANN Bench is versioned within the official CANN repository and is designed for long-term community co-construction, providing the Ascend ecosystem with a quantitative, reproducible, and sustainably maintained yardstick for AI operator-authoring capability.

cs.AI↗

arXiv2026

Multi-Scale Dequant: Eliminating Dequantization Bottleneck via Activation Decomposition for Efficient LLM Inference

Quantization is essential for efficient large language model (LLM) inference, yet the dequantization step-converting low-bit weights back to high-precision for matrix multiplication has become a critical bottleneck on modern AI accelerators. On architectures with decoupled compute units (e.g., Ascend NPUs), dequantization operations can consume more cycles than the matrix multiplication itself, leaving the high-throughput tensor cores underutilized. This paper presents Multi-Scale Dequant (MSD), a quantization framework that removes weight/KV dequantization from the GEMM critical path. Instead of lifting low-bit weights to BF16 precision, MSD decomposes high-precision BF16 activations into multiple low-precision components, each of which can be multiplied directly with quantized weights via native hardware-accelerated GEMM. This approach shifts the computational paradigm from precision conversion to multi-scale approximation, avoiding INT8-to-BF16 weight conversion before GEMM. We instantiate MSD for two weight formats and derive tight error bounds for each. For INT8 weights (W4A16), two-pass INT8 decomposition achieves near 16 effective bits. For MXFP4 weights (W4A16), two-pass MXFP4 decomposition yields near 6.6 effective bits with error bound 1/64 per block surpassing single-pass MXFP8(5.24 bits) while maintaining the same effective GEMM compute time. We further derive closed-form latency and HBM traffic models showing that MSD avoids the Vector-Cube pipeline stall caused by dequantization and reduces KV cache HBM traffic by up to 2.5 times in attention. Numerical simulations on matrix multiplication and Flash Attention kernels confirm that MSD does not degrade accuracy compared to dequantization baselines, and in many settings achieves lower L2 error.

stat.ML↗

arXiv2025

AMLA: MUL by ADD in FlashAttention Rescaling

Multi-head Latent Attention (MLA) significantly reduces KVCache memory usage in Large Language Models while introducing substantial computational overhead and intermediate variable expansion. This poses challenges for efficient hardware implementation -- especially during the decode phase. This paper introduces Ascend MLA (AMLA), a high-performance kernel specifically optimized for Huawei's Ascend NPUs. AMLA is built on two core innovations: (1) A novel FlashAttention-based algorithm that replaces floating-point multiplications with integer additions for output block rescaling, leveraging binary correspondence between FP32 and INT32 representations; (2) A Preload Pipeline strategy with hierarchical tiling that maximizes FLOPS utilization: the Preload Pipeline achieves Cube-bound performance, while hierarchical tiling overlaps data movement and computation within the Cube core. Experiments show that on Ascend 910 NPUs (integrated in CloudMatrix384), AMLA achieves up to 614 TFLOPS, reaching 86.8% of the theoretical maximum FLOPS, outperforming the state-of-the-art open-source FlashMLA implementation, whose FLOPS utilization is up to 66.7% on NVIDIA H800 SXM5. The AMLA kernel has been integrated into Huawei's CANN and will be released soon.

cs.LG↗

arXiv2023

Proof-of-Principle Demonstration of Fully-Passive Quantum Key Distribution

Quantum key distribution (QKD) offers information-theoretic security based on the fundamental laws of physics. However, device imperfections, such as those in active modulators, may introduce side-channel leakage, thus compromising practical security. Attempts to remove active modulation, including passive decoy intensities preparation and polarization encoding, have faced theoretical constraints and inadequate security verification, thus hindering the achievement of a fully passive QKD scheme. Recent research has systematically analyzed the security of a fully passive modulation protocol. Based on this, we utilize the gain-switching technique in combination with the post-selection scheme and perform a proof-of-principle demonstration of a fully passive quantum key distribution with polarization encoding at channel losses of 7.2 dB, 11.6 dB, and 16.7 dB. Our work demonstrates the feasibility of active-modulation-free QKD in polarization-encoded systems.

quant-ph↗

arXiv2022

Passive continuous variable quantum key distribution

Passive quantum key distribution (QKD) has been proposed for discrete variable protocols to eliminate side channels in the source. Unfortunately, the key rate of passive DV-QKD protocols suffers from sifting loss and additional quantum errors. In this work, we propose the general framework of passive continuous variable quantum key distribution. Rather surprisingly, we find that the passive source is a perfect candidate for the discrete-modulated continuous variable quantum key distribution (DMCV QKD) protocol. With the phase space remapping scheme, we show that passive DMCV QKD offers the same key rate as its active counterpart. Considering the important advantage of removing side channels that have plagued the active ones, passive DMCV QKD is a promising alternative. In addition, our protocol makes the system much simpler by allowing modulator-free quantum key distribution. Finally, we experimentally characterize the passive DMCV QKD source, thus showing its practicality.

quant-ph↗