设备优先反馈：面向移动端原生大语言模型驱动的神经网络架构搜索
Device-First Feedback: Toward Mobile-Native LLM-Driven Neural Architecture Search
- University of Würzburg（维尔茨堡大学）
- CAIDAS
- IFI
机构由 AI 辅助整理，请以论文原文为准。
AI总结：
该研究提出自动化移动端部署闭环流水线，针对CIFAR-10、CIFAR-100基准在三星平板验证，发现GPU微调无法保证移动端性能单调提升，需多数据集设备端测量。
AI中文摘要：
将大语言模型（LLM）生成的卷积神经网络部署到实际移动硬件上，仅靠GPU验证准确率是不够的：INT8格式的TensorFlow Lite导出、算子选择以及设备端延迟共同决定了模型是否可用。我们提出了一种自动化移动端部署流水线，形成从架构生成型LLM的QLoRA微调，到GPU评估、INT8导出、物理设备基准测试，再到训练语料库的门控增强的闭环。该流水线完全脚本化，可逐轮运行且无需人工干预，支持中断后恢复。我们在三星SM-P613平板（随机种子42，每轮20个模型，第0-6轮）上，针对CIFAR-10和CIFAR-100两个基准测试评估了相同的固定协议。在CIFAR-10上，第1轮通过门控，移动端部署得分较基线提升约25.6倍，量化准确率均值为46.9%；后续轮次提升了GPU准确率，但未通过非递减移动端门控。在CIFAR-100上，QLoRA前的基线保持最佳移动端得分；迭代轮次提升了GPU准确率（最高达26.2%），但无法超越第0轮的设备端表现，训练池在首次通过轮次后停滞在19个样本。两项研究共同表明，闭环GPU微调无法保证移动端性能的单调提升，尤其是在更复杂的分类任务中，需要多数据集、设备端测量来充分测试部署目标。我们发布了带95%置信区间的每轮指标、所有图表及完整复现命令。
英文摘要：
Deploying convolutional neural networks generated by large language models (LLMs) on real mobile hardware requires more than GPU validation accuracy: INT8 TensorFlow Lite export, delegate selection, and on-device latency jointly determine whether a model is usable. We present an automated mobile deployment pipeline that closes the loop from QLoRA fine-tuning of an architecture-generating LLM through GPU evaluation, INT8 export, and physical-device benchmarking to gated augmentation of the training corpus. The pipeline is fully scripted and runs cycle-by-cycle without manual intervention, with resume support after interruptions. We evaluate the same frozen protocol on two benchmarks, CIFAR-10 and CIFAR-100, on a Samsung SM-P613 tablet (seed 42, 20 models per cycle, cycles 0-6). On CIFAR-10, cycle 1 is gate-accepted and improves the mobile deployment score approximately 25.6x over the baseline with a mean quantized accuracy of 46.9%; later cycles raise GPU accuracy but fail the non-decreasing mobile gate. On CIFAR-100, the pre-QLoRA baseline retains the best mobile score; iterative rounds improve GPU accuracy (up to 26.2%) yet cannot surpass cycle 0 on-device, and the training pool stalls at 19 examples after the first accepted round. Together, the two studies show that closed-loop GPU fine-tuning does not guarantee monotonic mobile gains, especially on harder classification tasks, and that multi-dataset, on-device measurement is needed to stress-test deployment objectives. We release per-cycle metrics with 95% confidence intervals, all figures, and complete reproduction commands.
