Call for Speakers: PyTorch Day Korea 2026!

PyTorch Day Korea 2026 is the first large-scale offline conference hosted by the PyTorch Korea User Group.

We invite you to share your experience and knowledge in building, serving, and operating models using PyTorch in production.

We welcome not only success stories but also failed experiments, performance optimization journeys, technical trade-offs, and open-source contribution stories.

First-time speakers are also highly encouraged and welcome to submit proposals.

Event Information

• Event Name: PyTorch Day Korea 2026
• Date: Saturday, November 21, 2026
• Location: 18th Floor, Centerfield East · AWS Korea, Seoul
• Presentation Language: Korean or English
• Format: In-person (Offline)
• Official Website: pytorchday.kr

Timeline

• CfP Opens: Monday, July 13, 2026
• CfP Closes: Saturday, August 22, 2026, 23:59 KST
• Acceptance Notification: Monday, August 31, 2026
• Schedule Release: Monday, September 7, 2026
• Event Date: Saturday, October 31, 2026

Tracks

PyTorch Day Korea 2026 features three tracks—Build, Serve, and Run—following the lifecycle of how AI systems are created and delivered to users.

Please select the track that best aligns with the core technical contribution of your presentation. Note that the organizing committee may suggest a more suitable track during the program structuring phase.

01. Build Track

This track focuses on experiences and technologies related to building, training, and optimizing models. It covers PyTorch-based model development, research/development use cases, and model optimization.

Example Topics:
* Model development and research use cases using PyTorch
* PyTorch Core features and internal architecture
* torch.compile, TorchInductor, Custom Operator, and Kernel development
* Distributed training and large-scale model training
* Fine-tuning, PEFT, and training efficiency
* Model evaluation, benchmarking, reproducibility, and performance improvement

02. Serve Track

This track explores experiences and technologies in deploying and inferencing models using open-source AI projects within the PyTorch ecosystem. It covers the intersection of models and infrastructure, including model serving frameworks, inference engines, distributed inference, accelerator operations, and performance optimization using GPUs/NPUs, vLLM, DeepSpeed, Ray, Safetensors, etc.

Example Topics:
* Production deployment and serving architecture for PyTorch models
* Leveraging open-source projects such as vLLM, DeepSpeed, Ray, and Safetensors
* Distributed inference for LLMs and multimodal models
* Quantization, model lightweighting, batching, and KV cache optimization
* Inference systems built on GPUs, NPUs, and various accelerators
* Throughput, latency, cost optimization, and handling model server failures

03. Run Track

This track highlights experiences in building AI applications and services powered by PyTorch models and delivering them to end-users.

Example Topics:
* Generative AI-based application development
* AI Agents and tool-calling workflows
* RAG, search, and context/memory management
* Integrating AI capabilities into existing products and services
* AI application evaluation, quality control, and leveraging user feedback
* Observability, reliability, security, and cost management for production AI services

Session Formats

Session Presentation

A standard presentation to explain a single topic with sufficient technical context and real-world examples. There is no limit to the number of proposals you can submit, but a maximum of one session will be selected per speaker.

• Duration: 30 minutes in total
• Recommended Structure: 25 minutes for presentation, 5 minutes for Q&A

Lightning Talk

Ideal for topics with a tightly defined scope, such as small open-source projects, interesting experimental results, short technical tips, or lessons learned from failures.

• Duration: 5 minutes
• Number of Speakers: 1 person

Tutorial

An in-depth, hands-on, and practical presentation involving 1 to 5 speakers on a specific topic.

• Duration: 60 minutes
• Number of Speakers: 1–5 people

Proposal Writing Guide

Your session description will be used not only by the program committee reviewing the proposals but also by attendees deciding which sessions to join once selected. Please keep the following guidelines in mind:

• Write a title that clearly conveys the core topic of your presentation on its own.
• Clearly explain the problem you solved and the technical contribution of your presentation.
• Do not just list the technologies used; explain what specific content will actually be covered.
• Include the concrete takeaways or knowledge attendees will gain from your session.
• Keep it concise and write in complete sentences.
• Avoid exaggerated or overly promotional language.
• Avoid using internal company jargon or proprietary product names to explain the core concepts.
• Clearly state any prerequisites or prior knowledge required for the session.

Submissions close on 2026-08-31 23:59 (Asia/Seoul), 2 weeks from now.