Prologue When first entering the world of large models, due to gaps in foundational math knowledge such as linear algebra, probability theory, and information theory, it’s easy to get lost among numerous terms: logprob (log probability), likelihood, NLL (Negative Log Likelihood), cross entropy, perplexity. They frequently appear in various corners of papers and documentation, yet they all feel…
我是 2024 年初到一家大模型公司工作,之前一直在数据库、存储等 infra 行业工作,因此有些很粗浅的转行认知。很久没有在 b 站做分享了,这次靠直播强制开机,回答了大家一些问题,稍稍弥合一点信息差。本文对直播中提到的一些点的稍微规整一点的总结,并将一些我觉得不错的资料附在最后。 b 站直播: https://www.bilibili.com/video/BV1uckJBkEto
I joined an LLM company in early 2024, having previously worked in the infra industry (databases, storage, etc.), so I have some very basic insights on switching careers. I haven’t shared on Bilibili for a long time; this live stream forced me to get back in gear. I answered some of your questions and bridged a bit of the information gap. This post is a slightly more organized summary of some…
Since becoming distinctly self-aware, never have I clashed so intensely with the world and with myself as I have this year—yet the result is strangely magical: I have become even more peaceful. Many subconscious reactions, many habitual practices, when excavated inward, can be traced back to such ancient reinforcement chains. Just as Shi Tiesheng said—the bullet fired in youth strikes squarely…
Princeton COS 597R “Deep Dive into Large Language Models” 是普林斯顿大学的一门研究生课程,系统探讨了大语言模型原理、准备和训练、架构演进及其在多模态、对齐、工具使用等前沿方向中的应用与一些问题。注意,该课程侧重概念的理解上,而非工程的实现上。 我之前是在分布式系统和数据库内核方向,但这两年转到一家大模型公司做数据。本笔记主要是我对课程论文的梳理和精要。不同的是,我会结合在工作中解决实际问题的一些体感,给出一点转行人不同视角的思考,希望能对同样想从工程入门算法的同学一点帮助。 本文来自我的付费专栏《 系统日知录 》,欢迎订阅查看更多大模型解析文章,文末有优惠券信息。 本篇主要关注大模型的奠基之作——Transformer。 首先要明确问题域,Transformer…
Princeton COS 597R “Deep Dive into Large Language Models” is a graduate course at Princeton University that systematically explores the principles of large language models, their preparation and training, architectural evolution, and applications in cutting-edge directions such as multimodality, alignment, tool use, and related issues. Note that this course focuses on conceptual understanding…
As cloud infrastructure continues to mature, emerging companies typically move their infrastructure to the cloud in order to achieve business goals quickly. Developing on the cloud is actually quite different from traditional development using physical machines directly. The cloud emphasizes sharing and elasticity more, and as scale grows, isolation becomes important as well. These changes also…
streamlit is a Python library for quickly developing simple web apps. Its slogan is: A faster way to build and share data apps In other words, “a faster way to build and share data applications.” It is very popular in machine learning, data science, and even today’s large language model space. Its advantages are quite prominent: Uses the favorite language of developers in the above fields: Python.…
Introduction In certain workloads, memory usage gradually grows over time until an OOM occurs. Later, it was found to be a memory fragmentation issue. Replacing the system’s default memory allocator ( glibc malloc ) with jemalloc can effectively control the upper bound of memory growth. To understand the principles behind it, I sought out jemalloc’s original paper: A Scalable Concurrent malloc(3)…
Snowflake was founded in 2012 by two former Oracle employees, targeting cloud-native data warehouses from the very beginning. Therefore, its architectural design was (at the time) considered very “radical.” This forward-looking vision brought extraordinary returns — Snowflake went public in 2020 with a market capitalization reaching as high as $70 billion, setting the record for the largest…
I first heard about it from a passing mention on a podcast, so I sought out the audiobook for my commute. This is the 1939 translation by Fu Lei, with a faint, old-fashioned vernacular style. It’s a short book; I finished it in a few days. I like listening to things while walking—what enters my ears and what meets my eyes, the philosopher’s concise aphorisms and the myriad scenes of the street,…