Posts

This article explores three major categories of methods for accelerating LLM inference: model-level, computation-level, and decoding process. It provides a detailed introduction to the principles of speculative decoding and its equivalence proof, where a small model generates candidate sequences and a large model validates them, thereby improving inference speed.

Adam optimization method memory usage. Generally speaking, Adam uses more information, so it requires more memory. There was an experiment. When the Optimizer selects SGD, the Batch Size can be selected to be 20, but when it is replaced with Adam, the Batch Size can only be selected to be 2. The adaptive embodiment of Adam is exponential smoothing, which alleviates drastic changes in the gradient direction and makes the convergence path smoother.

This article introduces the file directory distribution of Linux systems, including `/bin`, `/sbin`, `/boot`, `/dev`, `/etc`, `/lib`, `/media`, `/mnt`, `/opt`, `/proc`, `/root`, `/run`, `/sys`, `/srv`, `/tmp`, `/usr`, `/var`, `/home`, etc.

This article introduces how to beautify in `powershell` under Windows, including modifying the theme, changing the background image and transparency, etc. The article first introduces how to install the .net version of `Powershell`, and then run specific commands to modify the theme and set the background image. Finally, the author emphasizes maintaining an understanding of basic DOM operations and not over-relying on modern front-end tools.

SEO experience learned from a certain article on Zhihu, though I can't recall which specific one it was.