2.0 -//Pentabarf//Schedule//EN PUBLISH BUTWTL@@pretalx.com -BUTWTL Opening Remarks by PyCon HK en en 20241116T100000 20241116T101000 001000

Opening Remarks by PyCon HK

PUBLIC CONFIRMED Short talk https://pretalx.com/pyconhk2024/talk/BUTWTL/ LT9 Scotty Kwok PUBLISH QVV8XV@@pretalx.com -QVV8XV Opening Remarks by CityU HK en en 20241116T101000 20241116T102000 001000

Opening Remarks by CityU HK

Opening Speech by College of Computing CityU HK PUBLIC CONFIRMED Lightning talk https://pretalx.com/pyconhk2024/talk/QVV8XV/ LT9 Dr. Ray Cheung PUBLISH XAFXDU@@pretalx.com -XAFXDU [Keynote] PyCon Hong Kong - the Story en en 20241116T102500 20241116T103000 000500

[Keynote] PyCon Hong Kong - the Story

PUBLIC CONFIRMED Lightning talk https://pretalx.com/pyconhk2024/talk/XAFXDU/ LT9 Sammy Fung PUBLISH W9X8DD@@pretalx.com -W9X8DD [Sponsored Keynote] Large Language Models Optimization with Python en en 20241116T103000 20241116T110000 003000

[Sponsored Keynote] Large Language Models Optimization with Python

The talk will delve into techniques and strategies for optimizing Large Language Models (LLMs) with Python. It focuses on addressing the computational challenges associated with training and deploying these models efficiently. One key aspect discussed is model parallelism, which involves distributing the model across multiple devices or instances to overcome memory limitations. Tensor parallelism, a form of model parallelism, is explored, where individual tensors are split across devices. Pipeline parallelism, another technique, enables concurrent execution of different model components on separate devices. The talk will also cover distributed training strategies, such as data parallelism and tensor-parallel language models, which leverage multiple devices to accelerate training. Techniques for reducing memory footprint, like quantization, pruning, and distillation, are explored as means to optimize LLM deployment. Optimizations for inference are also discussed, including model compression methods like quantization-aware training, pruning, and distillation. Kernel fusion, a technique that combines multiple operations into a single optimized kernel, is highlighted for improving inference performance. Additionally, the document explores accelerated inference using hardware accelerators. The talk aims to provide guidance on leveraging Python's capabilities for efficient LLM training, deployment, and inference. It covers a range of strategies and techniques to address the computational challenges associated with these models, enabling researchers and practitioners to optimize LLMs for improved performance and cost-effectiveness. PUBLIC CONFIRMED Talk https://pretalx.com/pyconhk2024/talk/W9X8DD/ LT9 Haowen Huang PUBLISH RUHCB8@@pretalx.com -RUHCB8 Sign and verify Python package with Sigstore keyless signing zh-hant zh-hant 20241116T111000 20241116T112500 001500

Sign and verify Python package with Sigstore keyless signing

PUBLIC CONFIRMED Short talk https://pretalx.com/pyconhk2024/talk/RUHCB8/ LT9 Frankie Ng PUBLISH NA7LPP@@pretalx.com -NA7LPP Network Automation for Improved Efficiency using Ansible and Python en en 20241116T112500 20241116T114000 001500

Network Automation for Improved Efficiency using Ansible and Python

This presentation explores how network automation can significantly enhance management efficiency. Traditional manual network configuration, provisioning, and maintenance are time-consuming, error-prone, and difficult to scale with network growth. Ansible, a popular open-source tool, is introduced as a key component for automating network configurations and workflows. We explain Ansible's agentless architecture and the use of playbooks—YAML files that define automated tasks for network devices. Through code examples and a sample playbook, we demonstrate how Python scripts can interact with network devices and be integrated into Ansible playbooks, enabling the automation of complex tasks efficiently. PUBLIC CONFIRMED Short talk https://pretalx.com/pyconhk2024/talk/NA7LPP/ LT9 TIMOTHY LAM PUBLISH YXQRCF@@pretalx.com -YXQRCF Build AI-powered RAG application with MySQL9.0 en en 20241116T115000 20241116T122000 003000

Build AI-powered RAG application with MySQL9.0

PUBLIC CONFIRMED Talk https://pretalx.com/pyconhk2024/talk/YXQRCF/ LT9 Ryan Kuan PUBLISH THDBMS@@pretalx.com -THDBMS Two roads diverged: the gap between web development and data science in Python en en 20241116T122500 20241116T124000 001500

Two roads diverged: the gap between web development and data science in Python

## Goal This talk aims to illustrate the differences and similarities between the Data Scientist and the Web Developer in Python. This talk will be of interest to an audience who is contemplating which path to take, or simply curious about how Python is used in different capacities. ## Outline - Part 1 (5 min): Mutual Myths. In this section I will discuss some common misconceptions about being a Data Scientist or Web Developer. - Part 2 (5 min): Transferrable Skills. In this section I will focus on the day-to-day responsibilities of the Web Developer, highlighting skills that are transferrable or learnable by the Data Scientist. - Part 3 (5 min): Recommendations. In this section, I will attempt to make some actionable recommendations for those contemplating to switch to the Software Engineering path in Python. PUBLIC CONFIRMED Short talk https://pretalx.com/pyconhk2024/talk/THDBMS/ LT9 Chan Sau Yee PUBLISH HN3F8L@@pretalx.com -HN3F8L End-to-end GPU Acceleration for scikit-learn and XGBoost en en 20241116T124000 20241116T125500 001500

End-to-end GPU Acceleration for scikit-learn and XGBoost

This talk will explore GPU acceleration beyond deep learning models and provide an overview of GPU-accelerated data science workflows. Python’s rich ecosystem has made it one of the most popular programming languages today. RAPIDS offers a suite of open-source Python libraries and primitives to accelerate core data science libraries, including pandas, scikit-learn, and NetworkX, without requiring any code changes. Additionally, the latest XGBoost integrates with RAPIDS to deliver a fully accelerated model training experience. We will demonstrate how to enable a GPU-accelerated end-to-end pipeline for training scikit-learn and XGBoost models, highlighting the significant speed improvements for various scikit-learn estimators. Then, we will delve into new features that facilitate scaling XGBoost using the latest NVIDIA Grace Hopper superchip to handle large datasets. We can discuss some details about the implementation and share our experience with GPU acceleration. Finally, we will outline our roadmap for future developments. PUBLIC CONFIRMED Short talk https://pretalx.com/pyconhk2024/talk/HN3F8L/ LT9 Jiaming Yuan PUBLISH PCZC3H@@pretalx.com -PCZC3H Python to Deploy Enterprise-Grade Delta Lake on AWS en en 20241116T140000 20241116T143000 003000

Python to Deploy Enterprise-Grade Delta Lake on AWS

PUBLIC CONFIRMED Talk https://pretalx.com/pyconhk2024/talk/PCZC3H/ LT9 Alan, Ka Hei Ng Jacky Kwok PUBLISH ES7DKX@@pretalx.com -ES7DKX Autonomous AI Agents for Dummies en en 20241116T144000 20241116T151000 003000

Autonomous AI Agents for Dummies

### Problem Statment Large Language Models (LLMs) like GPT-4 have several limitations that hinder their full potential. They often struggle with maintaining contextual understanding over extended conversations, leading to disjointed or repetitive interactions. Additionally, LLMs can lack accuracy, fail to provide real-time information, self-reasoning to decompose while planning the task. This is where Agents comes. Unlike traditional LLMs, AI agents are designed to self-reason and plan, mimicking human cognitive processes. They can interact with their environment, make decisions, and take actions autonomously. This capability enables them to overcome some of the contextual and reasoning challenges that LLMs face, making them more suitable for complex, dynamic tasks. ### My talk will cover: - How a simple execution of task is formed by humans? - Agentic Workflow and the major components required: This includes: Task, Memory, Tools, Agents, LLM and so on. - Planning and Reasoning: Under this I will cover Chain of thoughts, REACT, LATS prompt techniques, that is used to build Agentic workflow. - Conclusion PUBLIC CONFIRMED Talk https://pretalx.com/pyconhk2024/talk/ES7DKX/ LT9 Tarun Jain PUBLISH B9QPQF@@pretalx.com -B9QPQF Accelerating Python's performance with C and Cython en en 20241116T151500 20241116T153000 001500

Accelerating Python's performance with C and Cython

PUBLIC CONFIRMED Short talk https://pretalx.com/pyconhk2024/talk/B9QPQF/ LT9 Leo Chen PUBLISH TDYGRG@@pretalx.com -TDYGRG How to organize and deploy your Python applications with Docker en en 20241116T160000 20241116T163000 003000

How to organize and deploy your Python applications with Docker

PUBLIC CONFIRMED Talk https://pretalx.com/pyconhk2024/talk/TDYGRG/ LT9 Cristiano Pierandrei PUBLISH ZB8QGA@@pretalx.com -ZB8QGA Algorithmic Artistry: Musical Ideas for Pythonists en en 20241116T164000 20241116T171000 003000

Algorithmic Artistry: Musical Ideas for Pythonists

PUBLIC CONFIRMED Talk https://pretalx.com/pyconhk2024/talk/ZB8QGA/ LT9 Chuck-jee Chau PUBLISH YPKQZM@@pretalx.com -YPKQZM Data Validation in Python en en 20241116T172000 20241116T172500 000500

Data Validation in Python

PUBLIC CONFIRMED Lightning talk https://pretalx.com/pyconhk2024/talk/YPKQZM/ LT9 Meixin Wang PUBLISH 87VP87@@pretalx.com -87VP87 Async all the way: FastAPI and the ASGI era en en 20241116T172500 20241116T173000 000500

Async all the way: FastAPI and the ASGI era

PUBLIC CONFIRMED Lightning talk https://pretalx.com/pyconhk2024/talk/87VP87/ LT9 Taemin Ha PUBLISH 3RF9DP@@pretalx.com -3RF9DP Creative Problem Solving with Graphs en en 20241116T173000 20241116T173500 000500

Creative Problem Solving with Graphs

PUBLIC CONFIRMED Lightning talk https://pretalx.com/pyconhk2024/talk/3RF9DP/ LT9 Xiao Ying PUBLISH ZR3XP3@@pretalx.com -ZR3XP3 Interactive game with Mediapipe zh-hant zh-hant 20241116T173500 20241116T174000 000500

Interactive game with Mediapipe

PUBLIC CONFIRMED Lightning talk https://pretalx.com/pyconhk2024/talk/ZR3XP3/ LT9 Judy Wong PUBLISH UPTBUH@@pretalx.com -UPTBUH How many iPhones does it take to ship a new Python version? en en 20241116T174000 20241116T174500 000500

How many iPhones does it take to ship a new Python version?

PUBLIC CONFIRMED Lightning talk https://pretalx.com/pyconhk2024/talk/UPTBUH/ LT9 Chan Sau Yee PUBLISH JC3QG3@@pretalx.com -JC3QG3 匯豐咩價位? / How much is HSBC now? zh-hant zh-hant 20241116T174500 20241116T175000 000500

匯豐咩價位? / How much is HSBC now?

PUBLIC CONFIRMED Lightning talk https://pretalx.com/pyconhk2024/talk/JC3QG3/ LT9 Dr. Adrian Tam PUBLISH HTGZMQ@@pretalx.com -HTGZMQ Closing remarks en en 20241116T175500 20241116T181000 001500

Closing remarks

PUBLIC CONFIRMED Short talk https://pretalx.com/pyconhk2024/talk/HTGZMQ/ LT9 Scotty Kwok PUBLISH LY8XCV@@pretalx.com -LY8XCV Taiwan LLM: Bridging the Linguistic Divide with a Culturally Aligned Language Model en en 20241116T111000 20241116T114000 003000

Taiwan LLM: Bridging the Linguistic Divide with a Culturally Aligned Language Model

This will be an online presentation. In this presentation, we'll dive deep into the development and capabilities of TAIWAN-LLM, the first Large Language Model tailored for Traditional Chinese speakers in Taiwan. Key topics include: 1. The challenge of linguistic underrepresentation in existing LLMs 2. Our three-phase methodology: Continue-Pretraining, Supervised Fine-Tuning, and Feedback Supervised Fine-Tuning 3. The composition and curation of our Taiwanese corpus 4. Evaluation results on various NLP tasks, including contextual QA, summarization, and classification 5. Real-world applications and use cases of TAIWAN-LLM 6. The importance of culturally aligned language models for preserving linguistic diversity We'll also discuss the open-source release of TAIWAN-LLM and its potential impact on NLP research and applications for Traditional Chinese. PUBLIC CONFIRMED Talk https://pretalx.com/pyconhk2024/talk/LY8XCV/ LT8 Yenting Lin PUBLISH K7VYAH@@pretalx.com -K7VYAH Hackman: IoT & membership platform on Raspberry Pi & Nix for Hong Kong's first Hackerspace en en 20241116T115000 20241116T122000 003000

Hackman: IoT & membership platform on Raspberry Pi & Nix for Hong Kong's first Hackerspace

Hackman is a Hackerspace management system for Dim Sum Labs, Hong Kong's first Hackerspace. It is a Django application running on a Raspberry Pi 5 that manages our membership, controls access to the space, and much more. It serves a critical function for our space hence it must be reliable. Our members should easily add features to it while keeping the core stable. In this talk, we will talk about how we use various software engineering practices to make it stable and reliable while making it extensible. We'll discuss: * How the Django application is structured * How we use Redis as a means to easily extend the system * How we use Nix to make the operating system and the application easily reproducible * How we use DevOps/GitOps along with ample tests to manage software deployments Dim Sum Labs is Hong Kong’s first and longest-running Hackerspace since 2011, open to anyone interested in hacking: the intellectual challenge to creatively overcome or otherwise “hack” the limitations, capabilities, purposes, forms, etc. of virtually anything — or in other words: to mess around and build anything for fun. The members extend this ethos to its membership management system and the IoT in the space. For more information, visit us at https://www.dimsumlabs.com/ . Hackman is open source and can be accessed at https://github.com/dimsumlabs/hackman PUBLIC CONFIRMED Talk https://pretalx.com/pyconhk2024/talk/K7VYAH/ LT8 Nigel Choi PUBLISH HZAX7P@@pretalx.com -HZAX7P 如果HK Python User Group唔壯大，錦鯉就大鑊了! zh-hant zh-hant 20241116T122500 20241116T125500 003000

如果HK Python User Group唔壯大，錦鯉就大鑊了!

我會分享 過往超過十年 PYCON HK/ OSHK 的經驗，加左近年參加外地PYCON APAC之間的活動，去了解本地社群的不足，有更多事情可以改善，讓Python User Group 能夠有機地成長。包括 宣傳，設計，計劃，操作各範疇。 PUBLIC CONFIRMED Talk https://pretalx.com/pyconhk2024/talk/HZAX7P/ LT8 Calvin Tsang PUBLISH UGFV3S@@pretalx.com -UGFV3S Numbast: Bridging the gap between CUDA C++ and Python en en 20241116T140000 20241116T143000 003000

Numbast: Bridging the gap between CUDA C++ and Python

See Abstract. PUBLIC CONFIRMED Talk https://pretalx.com/pyconhk2024/talk/UGFV3S/ LT8 Michael Wang PUBLISH 9HHTWK@@pretalx.com -9HHTWK Operate with Confidence -- OpenTelemetry in Python en en 20241116T144000 20241116T151000 003000

Operate with Confidence -- OpenTelemetry in Python

This talk will be broken down into following sessions: - Introduction: 3 Pillars of System Observability (~5 mins) - OpenTelemetry on Python FastAPI and AWS Lambda, with Visualizations (~15 - 20 mins) - Why not just Logging? (~2 mins) - Why not just Metrics (with Prometheus)? (~2 mins) - Integration with other Cloud Monitoring Platforms (~2-3 mins) PUBLIC CONFIRMED Talk https://pretalx.com/pyconhk2024/talk/9HHTWK/ LT8 Alex Au PUBLISH GYCVTG@@pretalx.com -GYCVTG Leveraging Multi-Models and Open WebUI to Mimic ChatGPT with Data Security Considerations en en 20241116T151500 20241116T153000 001500

Leveraging Multi-Models and Open WebUI to Mimic ChatGPT with Data Security Considerations

The talk will cover the following aspects: ##### 1. Introduction to Open WebUI: ◦ Overview of Open WebUI, an extensible and feature-rich self-hosted WebUI designed to operate offline. ◦ Discussion on its capabilities to integrate various LLM runners, including Ollama and OpenAI-compatible APIs. ##### 2. Setting Up Open WebUI: ◦ Step-by-step guide on installing and configuring Open WebUI using Docker or Kubernetes for seamless deployment. ◦ Instructions on integrating GPU support for enhanced performance. ##### 3. Multi-Model Integration: ◦ Demonstration of how to leverage multiple models within Open WebUI, allowing for versatile and powerful interactions. ◦ Examples of using models such as LLaVA, Llama3, Phi-3 Mini, and more for diverse applications. ##### 4. Enhancing Functionality with Plugins: ◦ Introduction to the Pipelines Plugin Framework to incorporate custom logic and Python libraries. ◦ Examples of plugins for web search, document search, Discord integration, and more. ##### 5. Data Security and Control: ◦ Discussion on the importance of keeping data local or within company infrastructure. ◦ Best practices for ensuring data security and compliance while using in-house AI solutions. ##### 6. Building a Powerful Interface: ◦ Tips on extending Open WebUI to create a user interface similar to ChatGPT-4o. ◦ Leveraging features such as Markdown and LaTeX support, hands-free voice/video call, and retrieval-augmented generation (RAG) for a dynamic user experience. PUBLIC CONFIRMED Short talk https://pretalx.com/pyconhk2024/talk/GYCVTG/ LT8 Dr. Chung Ng PUBLISH UVKEGD@@pretalx.com -UVKEGD Time to Skip Tedious Steps - Spare Efforts with PyTorch Lightning en en 20241116T160000 20241116T163000 003000

Time to Skip Tedious Steps - Spare Efforts with PyTorch Lightning

PUBLIC CONFIRMED Talk https://pretalx.com/pyconhk2024/talk/UVKEGD/ LT8 Henry, Wai Yin Wong PUBLISH WQJV7B@@pretalx.com -WQJV7B Simplifying Python Web App Operations: Automating K8s Ops with Open Source en en 20241116T164000 20241116T171000 003000

Simplifying Python Web App Operations: Automating K8s Ops with Open Source

PUBLIC CONFIRMED Talk https://pretalx.com/pyconhk2024/talk/WQJV7B/ LT8 YangSoo Yoon PUBLISH FJLLGF@@pretalx.com -FJLLGF Pydantic Logfire: Empowering Python Observability en en 20241116T111000 20241116T114000 003000

Pydantic Logfire: Empowering Python Observability

PUBLIC CONFIRMED Talk https://pretalx.com/pyconhk2024/talk/FJLLGF/ LT7 Hemangi Karchalkar PUBLISH LNUWEC@@pretalx.com -LNUWEC High Throughput Python en en 20241116T115000 20241116T122000 003000

High Throughput Python

The talk is focus on CPython and experiment is done on Apple Silicon. It is to generate numpy array of random float as an example, and later to extend into Python lists of floats. I will compare the data generation throughput amongst using multiprocessing, threading, and concurrency.futures modules in Python as well as numba and joblib external libraries. The key result is to highlight the trade off between threading vs multiprocessing, in which if you want to use multiprocessing to work around the GIL, you pay the price of inter-process communication overhead. Even with free-threaded Python in 3.13 that you can avoid the GIL, you can't find a solution that is always better. PUBLIC CONFIRMED Talk https://pretalx.com/pyconhk2024/talk/LNUWEC/ LT7 Dr. Adrian Tam PUBLISH CEWPVB@@pretalx.com -CEWPVB Local 知識擂台LLM大格鬥 zh-hant zh-hant 20241116T122500 20241116T125500 003000

Local 知識擂台LLM大格鬥

This will be an online presentation. In the LLM world, every one is using GPT-4 as the golden standard. Are there cases where smaller language model can outperform GPT4 in terms of its richness in language expression and wealth of local knowledge? We want to present our discovery work on evaluating a suite of LLMs in the area of Cantonese skills, Hong Kong local geography and social knowledge. We will demo our open-sourced platform for others to play with the Cantonese, Hong Kong-specific chatbot arena as well. 一齊黎挑機啦！ PUBLIC CONFIRMED Talk https://pretalx.com/pyconhk2024/talk/CEWPVB/ LT7 Winnie Yeung Marcus Lau PUBLISH VRPMHX@@pretalx.com -VRPMHX How do I debug my PySpark workloads? en en 20241116T140000 20241116T143000 003000

How do I debug my PySpark workloads?

PUBLIC CONFIRMED Talk https://pretalx.com/pyconhk2024/talk/VRPMHX/ LT7 Hyukjin Kwon Allison Wang PUBLISH KZPSVD@@pretalx.com -KZPSVD Spark-less local data stack in 2024 en en 20241116T144000 20241116T151000 003000

Spark-less local data stack in 2024

PUBLIC CONFIRMED Talk https://pretalx.com/pyconhk2024/talk/KZPSVD/ LT7 Nok Lam Chan PUBLISH BHXJZA@@pretalx.com -BHXJZA Power PyTorch Training with Centralized AI Data Lake and Advanced Data Selection Techniques en en 20241116T151500 20241116T153000 001500

Power PyTorch Training with Centralized AI Data Lake and Advanced Data Selection Techniques

PUBLIC CONFIRMED Short talk https://pretalx.com/pyconhk2024/talk/BHXJZA/ LT7 Yang Cen PUBLISH X9UZSA@@pretalx.com -X9UZSA Enhancing Web Image Accessibility for Visually Impaired Individuals with Gemini Pro Vision and Google Cloud Platform zh-hant zh-hant 20241116T160000 20241116T163000 003000

Enhancing Web Image Accessibility for Visually Impaired Individuals with Gemini Pro Vision and Google Cloud Platform

# GeProVis AI Screen Reader GeProVis is an abbreviated term for Gemini Pro Vision, and my students have significantly enhanced the conventional Google ChromeVox Screen Reader by incorporating the robust capabilities of Google Gemini Pro Vision. This blog post will focus on the details of Google Cloud Platform (GCP). In brief, ChromeVox can extract the image source url and send it to GCP. In this talk, we will explain the technical details of the Python Google Cloud Function in this project. ### Tech Blog https://medium.com/google-developer-experts/enhancing-web-image-accessibility-for-visually-impaired-individuals-with-gemini-pro-vision-and-07190b97fc38 ### Story https://medium.com/google-developer-experts/hkiit-students-use-gemini-pro-vision-to-develop-ai-screen-reader-acace2a0f830 Hong Kong Google Cloud Summit 2024 GeProVis AI Screen Reader MD and GM, Google Hong Kong Michael Yue https://youtu.be/VqSYB62xrz8 ### Awards - City I&T Grand Challenge 2024 - Innovation Award - Google GDSC 2024 Solution Challenge Global Top 100 ### Speakers: Cyrus Wong - AWS ML Hero + Microsoft MVP - Azure AI + Google Developer Experts - GCP & AI/ML(GenAI) https://www.linkedin.com/in/cyruswong/ Hin Pak Markus Tsang (曾憲柏) - HKIIT 雲端系統及數據中心管理高級文憑課程 https://www.linkedin.com/in/hin-pak-markus-tsang-%E6%9B%BE%E6%86%B2%E6%9F%8F-327949b8/ Kelvin Yiu - AWS Cloud Club Captain & HKIIT 雲端系統及數據中心管理高級文憑課程 https://www.linkedin.com/in/kelvin-yiu-9a25b1290/ PUBLIC CONFIRMED Talk https://pretalx.com/pyconhk2024/talk/X9UZSA/ LT7 Cyrus Wong Markus Tsang YIU Kelvin