Jihao Xin

Post-training compression reduces LLM parameter counts but often produces irregular tensor dimensions that degrade GPU performance—a phenomenon we call dimensional misalignment. We present a full-stack analysis tracing root causes at three levels: framework, library, and hardware. The key insight is that model inference becomes slower because the resulting dimensions are unfriendly with the GPU execution stack. For example, compressing Llama-3-8B with activation-aware singular value decomposition (ASVD) has 15% fewer parameters yet runs no faster than the uncompressed baseline, because 95% of its dimensions are misaligned. We propose GAC (GPU-Aligned Compression), a new compression paradigm that wraps any dimension-reducing compressor and re-selects hardware-aligned dimensions via multi-choice knapsack optimization under the same parameter budget. We evaluate GAC on Llama-3-8B with ASVD and LLM-Pruner, achieving 100% alignment and recovering up to 1.5× speedup while preserving model quality.

In distributed training, communication often emerges as a bottleneck. In response, we introduce Kimad, a solution that offers adaptive gradient compression. By consistently monitoring bandwidth, Kimad refines compression ratios to match specific neural network layer requirements. Our exhaustive tests and proofs confirm Kimad's outstanding performance, establishing it as a benchmark in adaptive compression for distributed deep learning.

The shadows formed by fast-moving vehicles on a pavement PV array exhibit complex dynamic random distribution characteristics, which can cause a dynamic multipeak PV curve. Dynamic vehicle shadow will cause the reduction in pavement PV power, so the question is how to maximize the power in such conditions by operating at different maximum power point (MPP) quickly and continually. To address this issue, this paper proposes a maximum power point voltage forecasting method based on convolutional neural network (CNN). This method inputs the environmental information of pavement PV array into the proposed CNN model for learning and then uses this model to forecast the maximum power point voltage. Finally, simulation and experimental test with ResNet, MLP and CNN methods are carried out and the comparison results show that this model can accurately predict the maximum power point voltage of pavement PV array under different vehicle shading conditions.