Parse TPU pretraining profile data and document its schema.
npx skills add https://github.com/primatrix/skills --skill profile-anatomy
Please help me install this Agent Skill. Skill: profile-anatomy Source: https://github.com/primatrix/skills/tree/main/plugins/tpu-perf/skills/profile-anatomy Command: npx skills add https://github.com/primatrix/skills --skill profile-anatomy
This Skill provides schema documentation and reference scripts for interpreting TPU pretraining profile data, enabling deeper analysis and development of related tools.
Use the walk_xplane.py script to explore the full structure of a TPU pretraining profile directory.
High-intent search queries and answers about installing and using this skill.
TPU pretraining profiles contain structured performance and trace data. This Skill provides schema documentation and reference scripts to help developers understand the file layout, internal data types, and meaning within xplane.pb and trace.json.gz files.
You can parse TPU pretraining profile data by running the walk_xplane.py script. This script walks through the profile directory structure, reading xplane.pb and trace.json.gz files to extract and document their internal schema and contents.
Yes, you can analyze XLA profiles generated from TPU pretraining. The Skill specifically targets TPU pretraining profile data, providing the necessary schema documentation and parsing scripts to interpret XLA-related structures within the xplane.pb format.
Yes, you need protobuf dependencies to interpret TPU profile structures. The Skill relies on xplane_pb2 and hlo_pb2 protobuf definitions to parse the binary xplane.pb data and extract the underlying schema and trace information.
The best way to understand the internal schema of xplane.pb files is to use the provided schema documentation alongside the walk_xplane.py script, which systematically traverses and documents the file layout and data types within TPU profiles.
heikowagnerImplement Python tools for the NELA language toolchain.
lpjhelderDisplay agent session flow traces with chronological timelines and summaries.
doanchienthangdevOptimizes machine learning workloads on GPUs, TPUs, and edge accelerators with CUDA, TensorRT, and TFLite.
muskie9Build and analyze codebase dependency graphs with NetworkX.
