Me | Dezhi Yu

Efficient Cross-GPU Communication for Disaggregated LLM Serving

Fri, 12 Jun 2026 00:00:00 +0000

Direct Preference Optimization for Chatbot Fine-Tuning: An Empirical Study

Mon, 01 Jun 2026 00:00:00 +0000

Generative Modeling of Bach-Style Symbolic Music: A Comparative Study of Autoregressive, Latent-Variable, and Adversarial Approaches

Mon, 01 Jun 2026 00:00:00 +0000

How Far Are Video Models from True Multimodal Reasoning?

Wed, 01 Apr 2026 00:00:00 +0000

ROSE: A Reward-Oriented Data Selection Framework for LLM Task-Specific Instruction Tuning

Sat, 01 Nov 2025 00:00:00 +0000

DK-RRT: Deep Koopman RRT for Collision-Aware Motion Planning of Space Manipulators in Dynamic Debris Environments

Tue, 01 Jul 2025 00:00:00 +0000

AdaMixup: A Dynamic Defense Framework for Membership Inference Attack Mitigation

Wed, 01 Jan 2025 00:00:00 +0000

Deep Adaptive Control with Frequency Modulation for Aerospace Robotic Manipulators in Dynamic Object Transportation

Wed, 01 Jan 2025 00:00:00 +0000

KVDirect: Distributed Disaggregated LLM Inference

Wed, 01 Jan 2025 00:00:00 +0000

LaySummX at BioLaySumm: Retrieval-Augmented Fine-Tuning for Biomedical Lay Summarization Using Abstracts and Retrieved Full-Text Context

Wed, 01 Jan 2025 00:00:00 +0000

Machine Learning Optimizes the Efficiency of Picking and Packing in Automated Warehouse Robot Systems

Wed, 01 Jan 2025 00:00:00 +0000

Research on Reinforcement Learning Based Warehouse Robot Navigation Algorithm in Complex Warehouse Layout

Fri, 01 Nov 2024 00:00:00 +0000

Predicting 30-Day Hospital Readmission in Medicare Patients: Insights from an LSTM Deep Learning Model

Tue, 01 Oct 2024 00:00:00 +0000

Deep Reinforcement Learning-Based Obstacle Avoidance for Robot Movement in Warehouse Environments

Sun, 01 Sep 2024 00:00:00 +0000

LeetCode Cookbook

Thu, 13 Aug 2020 00:00:00 +0000

Segment Tree Basics

Wed, 15 Jan 2020 13:30:00 +0000

Redis Multi-Data Center Two-Way Synchronization

Thu, 31 Oct 2019 13:30:00 +0000

Redis Design Ideas and Usage Specifications

Tue, 18 Jun 2019 13:30:00 +0000

Detailed HTTP/2 header compression algorithm-HPACK

Sun, 16 Jun 2019 10:05:36 +0000

在 HTTP/1.1（请参阅）中，header 字段未被压缩。随着网页内的请求数增长到需要数十到数百个请求的时候，这些请求中的冗余 header 字段不必要地消耗了带宽，从而显着增加了延迟。

最初通过使用格式压缩 header 字段来解决此冗余问题，事实证明，这种格式非常有效地表示了冗余 header 字段。但是，这种方法暴露了安全风险，如 CRIME（轻松实现压缩率信息泄漏）攻击所证明的安全风险（请参阅）。

本规范定义了 HPACK，这是一种新的压缩方法，它消除了多余的 header 字段，将漏洞限制到已知的安全攻击，并且在受限的环境中具有有限的内存需求。介绍了 HPACK 的潜在安全问题。

HPACK 格式特意被设计成简单且不灵活的形式。两种特性都降低了由于实现错误而引起的互操作性或安全性问题的风险。没有定义扩展机制；只能通过定义完整的替换来更改格式。

点击看全文。

Golang Message Streaming Practice in Eleme

Fri, 15 Mar 2019 13:30:00 +0000

SQL Practical Optimization

Tue, 25 Dec 2018 13:30:00 +0000

How to understand gradient descent?

Sun, 21 Oct 2018 10:05:36 +0000

Gradient descent is a first-order iterative optimization algorithm for finding a local minimum of a differentiable function. To find a local minimum of a function using gradient descent, we take steps proportional to the negative of the gradient (or approximate gradient) of the function at the current point. But if we instead take steps proportional to the positive of the gradient, we approach a local maximum of that function; the procedure is then known as gradient ascent. Gradient descent is generally attributed to Cauchy, who first suggested it in 1847, but its convergence properties for non-linear optimization problems were first studied by Haskell Curry in 1944.

Click to read full article.

TLS 1.3 Handshake Protocol

Sun, 21 Oct 2018 10:05:36 +0000

握手协议用于协商连接的安全参数。握手消息被提供给 TLS 记录层，在记录层它们被封装到一个或多个 TLSPlaintext 或 TLSCiphertext 中，它们按照当前活动连接状态进行处理和传输。

 enum {
 client_hello(1),
 server_hello(2),
 new_session_ticket(4),
 end_of_early_data(5),
 encrypted_extensions(8),
 certificate(11),
 certificate_request(13),
 certificate_verify(15),
 finished(20),
 key_update(24),
 message_hash(254),
 (255)
 } HandshakeType;

 struct {
 HandshakeType msg_type; /* handshake type */
 uint24 length; /* remaining bytes in message */
 select (Handshake.msg_type) {
 case client_hello: ClientHello;
 case server_hello: ServerHello;
 case end_of_early_data: EndOfEarlyData;
 case encrypted_extensions: EncryptedExtensions;
 case certificate_request: CertificateRequest;
 case certificate: Certificate;
 case certificate_verify: CertificateVerify;
 case finished: Finished;
 case new_session_ticket: NewSessionTicket;
 case key_update: KeyUpdate;
 };
 } Handshake;

协议消息必须按照一定顺序发送(顺序见下文)。如果对端发现收到的握手消息顺序不对，必须使用 “unexpected_message” alert 消息来中止握手。

点击看全文。

Fundamentals of Cryptography

Sun, 15 Jul 2018 13:30:00 +0000

The Practice of Spatial Index in Geographic Service

Sun, 14 Jan 2018 13:30:00 +0000

Getting started with Machine Learning

Wed, 08 Nov 2017 13:30:00 +0000

How to design and implement a thread-safe Map data structure

Sun, 10 Sep 2017 10:05:36 +0000

Map 是一种很常见的数据结构，用于存储一些无序的键值对。在主流的编程语言中，默认就自带它的实现。C、C++ 中的 STL 就实现了 Map，JavaScript 中也有 Map，Java 中有 HashMap，Swift 和 Python 中有 Dictionary，Go 中有 Map，Objective-C 中有 NSDictionary、NSMutableDictionary。

上面这些 Map 都是线程安全的么？答案是否定的，并非全是线程安全的。那如何能实现一个线程安全的 Map 呢？想回答这个问题，需要先从如何实现一个 Map 说起。

点击看全文。

iOS Master Book WWDC 2017

Fri, 01 Sep 2017 00:00:00 +0000

Functional Reactive Programming Thinking

Thu, 15 Jun 2017 13:30:00 +0000

iOS Master Book Summer

Thu, 01 Jun 2017 00:00:00 +0000

Eleme Quarterly Newcomer Report

Tue, 18 Apr 2017 13:30:00 +0000

Weex layout engine powered by FlexBox algorithm

Fri, 31 Mar 2017 10:05:36 +0000

在上篇文章里面谈了Weex在iOS客户端工作的基本流程。这篇文章将会详细的分析Weex是如何高性能的布局原生界面的，之后还会与现有的布局方法进行对比，看看Weex的布局性能究竟如何。

打开Weex的源码的Layout文件夹，就会看到两个c的文件，这两个文件就是今天要谈的Weex的布局引擎。

Layout.h和Layout.c最开始是来自于React-Native里面的代码。也就是说Weex和React-Native的布局引擎都是同一套代码。

当前React-Native的代码里面已经没有这两个文件了，而是换成了Yoga。

点击看全文。

iOS Master Book Spring

Wed, 01 Mar 2017 00:00:00 +0000

iOS Architecture Smalltalk

Thu, 25 Aug 2016 13:30:00 +0000

Phabricator Introduce

Wed, 20 Jan 2016 13:30:00 +0000