Model Explorer を使用して、量子化の影響を受ける問題のあるオペレーションを特定します。オペレーションをエラー指標で並べ替えて品質の低下を見つけ、レイヤごとの分析情報を取得し、さまざまな量子化結果を比較して、理想的なモデルサイズと品質のトレードオフを見つけます。
Model Explorer を使用すると、ベンチマークとデバッグ ツールの出力をより詳しく把握できます。GPU で実行できるオペレーションに関する分析情報を取得し、オペレーションをレイテンシで並べ替え、アクセラレータ間でオペレーションごとのパフォーマンスを比較します。
大規模なモデルのサポート
Model Explorer は、大規模なモデルをシームレスにレンダリングするように設計されています。数千のノード？、GPU ベースのレンダリング エンジンは、非常に大きなモデルでもスムーズにレンダリングできるようにスケールアップできます。また、Model Explorer の独自のアプローチにより、ファイルとフォルダのシステムのようにレイヤを閉じることができるため、より迅速で簡単な操作が可能になります。
作業を効率化するための機能
Search
分割ビュー
データ オーバーレイ
強力な正規表現ベースの検索により、特定のノードの検索、フィルタ、ハイライト表示が可能です。
同じタブにモデルを並べて読み込み、簡単に比較できます。
ノード固有のカスタムデータを Model Explorer に読み込んで、モデルのホットスポットなどの問題をすばやく特定します。
[[["わかりやすい","easyToUnderstand","thumb-up"],["問題の解決に役立った","solvedMyProblem","thumb-up"],["その他","otherUp","thumb-up"]],[["必要な情報がない","missingTheInformationINeed","thumb-down"],["複雑すぎる / 手順が多すぎる","tooComplicatedTooManySteps","thumb-down"],["最新ではない","outOfDate","thumb-down"],["翻訳に関する問題","translationIssue","thumb-down"],["サンプル / コードに問題がある","samplesCodeIssue","thumb-down"],["その他","otherDown","thumb-down"]],["最終更新日 2025-01-13 UTC。"],[],[],null,["# Model Explorer\n\n**A visualization tool that lets you analyze ML models and graphs, accelerating\ndeployment to on-device targets.**\n\n[Get Started](https://github.com/google-ai-edge/model-explorer/wiki/1.-Installation)\n[Try it in Colab](https://github.com/google-ai-edge/model-explorer/blob/main/example_colabs/quick_start.ipynb)\n[Learn More](https://github.com/google-ai-edge/model-explorer/)\n\n### Making edge development faster\n\nThe constraints of edge devices often necessitate extra steps to convert and\noptimize models before they run efficiently, and visualization is one of the\nmost effective ways to understand a model and identify targets for optimization.\n\n| **Conversion** | **Quantization** | **Optimization** |\n|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|\n| Model Explorer's side-by-side comparison feature makes it easier to spot conversion-related issues. Navigate the graph layer by layer, diving deeper into the graph by expanding and collapsing sections. Inspect the internal structure and connections within graphs at the granularity you need. | Use Model Explorer to identify problematic operations affected by quantization. Sort ops by error metrics to find quality drops, get insights per layer, and compare different quantization results to find the ideal model size-quality trade-off. | Use Model Explorer to better understand the output from your benchmarking and debugging tools. Gain insights into which ops can run on GPU, sort ops by latency, and compare per-op performance across accelerators. |\n\n### Support for large models\n\nModel Explorer is designed to render large models seamlessly. Thousands of\nnodes? No problem. The GPU-based rendering engine is capable of scaling up to\nsmoothly render even very large models. And Model Explorer's unique approach to\ncollapsing layers like a system of files and folders means that it's faster and\neasier to navigate.\n\n### Features designed to help you work faster\n\n| **Search** | **Split View** | **Data Overlays** |\n|-------------------------------------------------------------------------------------|--------------------------------------------------------------------------------------------|---------------------------------------------------------------------------------------------------------------------|\n| Powerful regex-based search helps you locate, filter, and highlight specific nodes. | Load models side by side in the same tab for easy comparison. | Load custom, node-specific data into Model Explorer to quickly identify hot spots and other issues with your model. |\n| With the click of a button, export an image of the graph to share with your team. | Save your location in the graph by adding bookmarks, making it easy to jump between areas. | View tensor shapes, trace inputs and outputs, highlight identical layers, see child node counts, and more. |\n\n### Two ways to use Model Explorer\n\n| **Run it locally** | **Run it in a Colab notebook** |\n|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|\n| Follow the easy [installation instructions](https://github.com/google-ai-edge/model-explorer/wiki/1.-Installation) on GitHub to set up Model Explorer on your local machine. It runs in a browser window and all your data stays local. Supports Linux, Mac and Windows. | Model Explorer runs well in Colab, meaning you can integrate it into your existing model development workflow. Try the [demo notebook](https://github.com/google-ai-edge/model-explorer/blob/main/example_colabs/quick_start.ipynb) or follow the [installation instructions](https://github.com/google-ai-edge/model-explorer/wiki/5.-Run-in-Colab-Notebook) to add it to your own. |"]]
