InGoogle Developer ExpertsbyPark Chansung·1d agoPyTorch × TPU (TorchTPU & OpenXLA): Compiler Internals and Operator Fusion OptimizationTL;DR — Real TPU throughput comes from minimizing memory round-trips: TorchTPU’s OpenXLA compiler captures the full graph and fuses…
Park Chansung·2d agoPyTorch × TPU (TorchTPU & OpenXLA): 컴파일러 동작 원리와 연산 융합 최적화TL;DR — TPU 성능의 핵심은 메모리 왕복 최소화에 있으며, TorchTPU의 OpenXLA 컴파일러는 전체 연산 그래프를 온칩 메모리 안에서 하나로 융합해 이를 달성한다. 단, 이 이점을 유지하려면 CPU 동기화로 인한 그래프 분할과 가변…
Park Chansung·Jul 10Getting the Most Out of RTX PRO 6000H100 is expensive. RTX PRO 6000 Blackwell is cheaper on JarvisLabs, it is a newer Blackwell GPU, and it supports NVFP4. NVFP4 is a 4-bit…
Park Chansung·Jul 9Automating the LLM Fine-Tuning Lifecycle on GCP TPU v5e via Multi-Agent WorkflowsAutomating the LLM Fine-Tuning Lifecycle on GCP TPU v5e via Multi-Agent Workflows
Park Chansung·Jun 17Always-On OpenClaw, On-Demand InferenceOpen-weight models are attractive for personal assistants because they offer control and avoid depending on a hosted API for every answer…
Park Chansung·Jun 15RTX PRO 6000 for AI Research: When Is It Enough?RTX PRO 6000 is not an H100 killer. It is a practical research GPU when the workload fits, the iteration loop matters, and H100 is not yet…
Park Chansung·Apr 30Building a Diffusion Model Research Harness from Scratch with Keras 3, JAX, and Google Cloud TPUI wanted to build a diffusion model — from scratch, not from a tutorial — that could generate realistic fashion item images. Then I wanted…
Park Chansung·Nov 29, 2025Automating Academic Poster Generation with arxiv2poster: From Research Papers to Visual…Academic poster creation has long been a time-consuming task for researchers. From designing layouts to selecting key information, the…
Park Chansung·Nov 19, 2025Seamless AI Dev: Bridging Local VSCode and Google ColabFor developers and data scientists, the choice between a local environment and cloud resources often feels like a trade-off. You love the…
InGoogle Developer ExpertsbyPark Chansung·Feb 24, 2025Distilling from Dialogues: Finding Meaning in LLM InteractionsIn the age of Large Language Models (LLMs), engaging in conversations with AI has become increasingly common. However, as these…