Low-Rank Decomposition
A model compression technique that decomposes a large matrix into the product of several smaller matrices to reduce parameter count and computation. It is commonly used to perform low-rank approximation of the recurrent state in state space models.
Timeline (last 90 days)
Model acceleration aims to reduce the number of sampling steps, and typical methods include DDIM and distillation methods.
Quantization, pruning, and low-rank decomposition are three common means of model compression.
Low-rank decomposition introduces truncation errors that degrade model accuracy, which must be compensated through brief fine-tuning (referred to as healing/repair).
All Facts (3)
Model acceleration aims to reduce the number of sampling steps, and typical methods include DDIM and distillation methods.
60%UnverifiedQuantization, pruning, and low-rank decomposition are three common means of model compression.
50%UnverifiedLow-rank decomposition introduces truncation errors that degrade model accuracy, which must be compensated through brief fine-tuning (referred to as healing/repair).
50%