Tag: inference efficiency

A boundary-aware U-Net and adaptive Hough transform framework achieves high-accuracy static feature extraction from flotation froth images.

A benchmark across 6,900 outputs revealed that reasoning modes substantially increased clinical text extraction accuracy while noticeably lengthening inference latency.

A controlled benchmark study finds that 4-bit quantization of a compact Transformer model delivers the strongest overall efficiency gains, while ...

A new conditional diffusion framework called CDJMP predicts multi-agent vehicle trajectories with state-of-the-art accuracy while cutting inference time to under ...

In a controlled comparison on one RTX 4090, attention pooling reduced batch-one GPU memory use by 62.0% while maintaining comparable ...

A new survey charts how discrete diffusion models have evolved from theoretical curiosities into billion-parameter systems that can match or ...