AI Systems DevLabs Summer 2026

The AI Systems DevLabs Summer 2026 in Sunnyvale!

Date: June 23rd and June 24th, 2026
Location: Sunnyvale, CA

Recordings from this event can be found in the AI Systems DevLabs Summer 2026 playlist on the OpenXLA YouTube channel.

Event Description

This event offers a casual, hands-on opportunity to learn about development, meet fellow members of the community, and deep dive into specific needs with engineers from Google.

Schedule

Please see the Full Schedule for session details.

Recordings

Event Recordings Playlist on YouTube

# Title Link
1 Pallas tutorial, DevLabs 2026 Recording
2 JAX: What's New Recording
3 Performance Optimization with Xprof Recording
4 Tokamax: A library of High-Performance Custom Kernels Recording
5 Grain: Data Processing for JAX models Recording
6 Ragged Paged Attention: A High-Performance and Flexible LLM Inference Kernel for TPU Recording
7 FastAFD: Open-Source Large-Scale Attention–FFN Disaggregation for Efficient MoE Serving Recording
8 Benchmarking vLLM on XLA Backends: From CUDA to OpenXLA for LLM Serving Recording
9 NNX v2: Simplifying Neural Net APIs Recording
10 Fireside Chat with Jeff Dean Recording
11 UniDFlow: A Unified Discrete Flow Framework for Multimodal Reasoning and Generation Recording
12 Pallas Compiler: Mosaic Recording
13 Scaling Pretrained Representations Enables Label-Free Out-of-Distribution Detection w/o Fine-Tuning Recording
14 Hands On Workshop: Profiling & Debugging JAX/OpenXLA Workloads Recording
15 Pallas kernels on TPU SparseCore Recording
16 Enhancing Medical Knowledge in LLMs via Supervised Continued Pretraining on Clinical Notes Recording
17 MaxText for Large Scale Pre-Training Recording
18 TorchTPU: Running PyTorch Natively on TPUs Recording
19 Using Motion Grids and On-Device Gemma to Translate Real-Time Indonesian Sign Language Recording
20 Agents for ML Dev - MaxCode/MaxKernel Recording
21 Tensor-Based Relational Query Execution on Google TPUs Recording
22 JAXQuantum: accelerating quantum hardware simulation with TPUs Recording
23 Kinetic: Zero-Ops ML Scaling for JAX and Keras Recording
24 Joint Sharding and Placement for Video Generation on TPUs Recording
25 From Triton to Pallas: Agentic, Evolutionary Kernel Porting to JAX Recording
26 MaxText: Reinforcement Learning at Frontier Scale on Google Cloud Recording
27 XLA Crash Course Recording
28 Integrating FlyDSL into JAX/XLA on AMD Recording
29 Host Offloading for Faster and Cost-Efficient LLM Training with TPU and Intel Xeon® Processor Recording
30 oneAPI backend for Intel GPUs Recording
31 Architecting Large Scale Agentic RL Workflow on TPU Recording
32 JAX-AITER: Bringing AMD ROCm AI Kernels into JAX/XLA Recording
33 Mastering LLM Post Training with Tunix Recording
34 Porting QwenImage to PyTorch XLA: From GPU to TPU with xprof-Guided Optimization Recording
35 Scaling LLM Inference on TPUs: Recent Developements, Optimizations, and RL Integrations Recording
36 Fusion Device Shape Optimization with Finite Elements in JAX Recording
37 Customizing JAX stack with NVIDIA's kernel DSLs Recording
38 Accelerating JAX Inference on GPUs with FlashInfer Kernels Recording
39 Scaling Next-Generation AI Workloads: Cloud TPU Architecture and the OSS Stack Recording
40 How to scale models in JAX on NVIDIA GPU Recording
41 The Evolution of Kubernetes: Building the Distributed OS for AI Recording
42 Leveraging offloading in JAX on NVIDIA GPU Recording
43 Checkpointing with Orbax: Unlocking Performance at Scale for RL Workloads. Recording
44 Helion on TPU: expand the Helion DSL to automatically emit accurate and fast TPU kernels Recording
45 From Scratch to Silicon: Recreating TPU v1 and the Future of Open-Source AI Systems Education Recording
46 Evolving the Ray ecosystem for TPUs Recording
47 SGLang on TPUs: Production-Grade, High-Performance LLM Serving with PyTorch and JAX Recording
48 TPU SparseCore: Hardware and Software Recording
49 Thought bubbles: an Unsupervised Method for Parallel Thinking in Latent Space Recording
50 Beyond the Chip: Re-Architecting the Ultimate AI Hypercomputer Recording
51 Reinventing Warehouse Computing for ML Recording
52 Get Started With TPU: From Zero to Hero Recording

Questions? Comments? We want to hear from you. You can email us at devlab@google.com