2 results about "Weight statistics" patented technology

The invention discloses a flood forecast evaluation method adapted to reservoir dispatching characteristics, which comprises the following steps: respectively driving the same hydrological model by using historical observation rainfall and forecast rainfall, quantifying the error contribution of the hydrological model and rainfall forecast through double-track difference, and constructing a sample evaluation weight in combination with a rainfall score; constructing a weighting condition error distribution model based on the weight, and generating a reservoir flow forecasting scene set containing physical cause characteristics; and inputting the scene set into a reservoir scheduling model containing flood control and benefit-making rules, and outputting a comprehensive scheduling risk index containing a downstream over-alarm risk and water abandoning loss through water balance and downstream evolution calculation. According to the method, the statistical weight is corrected through error attribution, the abstract forecast error is converted into specific engineering risk and economic loss, the hidden risk working condition which is difficult to find by a traditional statistical index is effectively identified, and a quantitative basis is provided for risk decision making of reservoir dispatching and targeted optimization of a forecast system.

This invention relates to a method for weight quantization of large language models based on statistical distribution, belonging to the field of artificial intelligence technology. This method adaptively selects the quantization threshold by analyzing the skewness, kurtosis, and dynamic range of the statistical weight distribution, achieving high-precision sparse preservation of important weights and low-bit compression of minor weights. Simultaneously, it introduces mean error compensation and variance alignment mechanisms, as well as a block-level parallel quantization strategy, to further reduce quantization errors and improve efficiency. This invention addresses the challenges of existing large language models with their massive parameter scale, requiring high-precision computation and substantial resources for inference, making them difficult to deploy on resource-constrained edge devices; and traditional quantization compression techniques suffer from significant accuracy loss, high computational overhead, and poor adaptability. This method achieves 8-bit and 4-bit quantization without retraining, achieving performance close to FP16 accuracy, significantly reducing model storage space and computational resource consumption, making it suitable for efficient deployment of large language models at edge devices.