Tag
Part 1 of Amazon's tutorial series on profiling TorchNeuron workloads on AWS Trainium, covering how to configure torch.profiler with the Neuron backend, read key_averages() output, and trace operations through Neuron Explorer.
HF Viewer 创作者宣布推出 embedl.com，一个收录 1000+ 针对边缘设备优化模型的平台，用户可通过免费的真实边缘设备农场（如 NVIDIA Orin、AGX Thor、RTX GPU）直接运行模型并获得详细性能分析。
Python 3.15 正式发布，Simon Willison 介绍了其中的重点特性：PEP 810 显式惰性导入（lazy imports）、PEP 798 列表推导式解包语法，以及新加入标准库的 Tachyon 采样性能分析器 profiling.sampling。
Simon Willison profiles his new web-edit app using Python 3.15's built-in sampling profiler, revealing that mixed line-ending preservation dominated CPU time, and demonstrates how to generate flame graphs and source heatmaps with the new tooling.
culpert 是一个实验性的 Rust 堆分配分析器，能将采样到的分配归属到 span，导出 pprof 兼容报告并在 CI 中检测分配回归，由 Cloudflare 开发并已在生产前捕获多个内存泄漏问题。
Python 3.15 is released, featuring lazy imports via a new `lazy` keyword, the Tachyon statistical sampling profiler, sentinel types, and more, alongside updated Astral tools Ruff, uv, and ty.
PyTorch Conference North America will feature a session by AWS's Esha Lakhotia and Pinak Panigrahi on profiling PyTorch workloads on AWS Trainium, covering torch.profiler capture through cycle-level hardware debugging with AI-assisted analysis. The event takes place in San Jose on October 20-21.
SparkTrace LLM 是一个新基准，用 Blackwell GPU 的真实计时结果评估 LLM 阅读 Nsight Compute 性能剖析输出后，能否正确选择最快的 CUDA kernel 重写版本。实验发现包括 GPT 5.4 nano 在内的一些模型会被性能剖析信息误导而选错 kernel。
Suvaditya Mukherjee of Hugging Face will present on using the PyTorch Profiler to identify performance bottlenecks and optimize model execution at PyTorch Conference North America 2026 in San Jose, October 20-21.
A free, open-source book titled 'How to Make Your Model Fast' that provides a systems view of efficient machine learning, from hardware to agents, to help optimize model performance.
The article discusses methods to avoid correctness space leaks in Haskell by classifying leaks into strictness, liveness, and sharing types, and recommends defensive coding patterns and profiling techniques.
Go 1.27 introduces a goroutine leak profiler to accurately detect and prevent goroutine leaks in running programs, including production systems, with minimal false positives.
The article investigates why Arrays.fill is 265 times slower on G1GC compared to ParallelGC, tracing the issue to JIT-generated code and memory barriers through detailed profiling and assembly analysis.
Aiki Alpha 3 is released with substantial performance improvements, including cheaper runtime realization, adaptive number representations, enhanced profiling and coverage, and stricter library constraints.
FleetSieve introduces a decision-critical profiling method for SLO-aware LLM fleet configuration that optimizes resource allocation by reducing unnecessary measurements, achieving efficiency gains over uniform profiling.
A software engineer reflects on how scientists often lack software engineering skills, using an example of optimizing an astrophysics simulation postprocessing tool and advocating for a 'missing semester' for scientists.
An open-source tool called XProf Cubism Reducer reduces TensorBoard trace sizes by over 90% for JAX/XLA, making performance profiling more efficient.
This article demonstrates how to profile eBPF code performance by creating a simple C test harness that measures file open latency, enabling developers to compare overhead before and after attaching eBPF hooks.
Two official PyTorch tutorials linking to the 'Profiling in PyTorch' series are highlighted, providing guidance on using the PyTorch profiler API for performance debugging.
CodeSizer is a static code size profiling tool for embedded firmware that uses objdump and addr2line to attribute code size to inline call trees and generates an HTML report.