trt-perf-analysis

Validate and analyze TensorRT performance data from paired JSON files.

13.2k|2.4k|Updated May 2, 2019

Search Tags:#optimization #performance analysis #tensorrt #backend comparison #layer profiling

Ready to run

npx skills add https://github.com/NVIDIA/TensorRT --skill trt-perf-analysis

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: trt-perf-analysis
Source: https://github.com/NVIDIA/TensorRT/tree/main/.agents/skills/trt-perf-analysis
Command: npx skills add https://github.com/NVIDIA/TensorRT --skill trt-perf-analysis

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

This Skill provides a comprehensive analysis and validation of TensorRT performance data, enabling users to understand and optimize their models efficiently.

Core Features & Use Cases

Performance Analysis: Validate and analyze TensorRT performance data from paired layer-info JSON and profile/latency JSON files.
Model Validation: Verify that layer/profile JSON files are valid and from the same model.
Basic Model Information: Infer basic model information from the analysis.
Optimization Opportunities: Identify fusion or latency optimization opportunities.
Markdown Report: Generate a concise Markdown performance report.
JSON Data: Provide structured JSON data for further analysis.
Use Case: Use this Skill when you need to inspect TensorRT performance reports, verify data integrity, and identify areas for optimization.

Quick Start

Run the analyze_trt_perf.py script with the input folder path: ./scripts/analyze_trt_perf.py /path/to/input/folder

Frequently Asked Questions about trt-perf-analysis

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I analyze TensorRT performance data from layer and profile JSON files?▼

You validate paired layer-info and profile/latency JSON files to analyze TensorRT performance data, identifying backend comparisons and optimization opportunities. The Skill processes these inputs to generate structured JSON and Markdown reports.

What is TensorRT layer profiling and how does it identify optimization opportunities?▼

TensorRT layer profiling inspects individual layer latencies from profile JSON files to identify optimization opportunities like layer fusion. Validating paired profile data pinpoints specific layers causing performance bottlenecks.

How do I validate that TensorRT profile and layer JSON files are from the same model?▼

You validate that TensorRT profile and layer JSON files are from the same model by analyzing the paired files for data integrity. This process verifies that input layer-info and profile/latency JSON files match structurally.

Does TensorRT performance analysis support backend performance comparison?▼

Yes, TensorRT performance analysis supports backend performance comparison by evaluating paired JSON files. It validates profile data to compare latencies across different backends and identify optimal configurations.

How do I generate a Markdown report for TensorRT performance data?▼

You generate a Markdown report for TensorRT performance data by running the analysis script on a folder containing paired layer and profile JSON files. The script outputs a concise Markdown summary alongside structured JSON data.

What format does TensorRT performance analysis output for further processing?▼

TensorRT performance analysis outputs structured JSON data for further processing and downstream pipelines. Alongside this JSON data, it generates a concise Markdown performance report for quick human review.