https://flagos-ai.github.io/FlagGems/performance/Recent content in Performance Benchmark on FlagGemsHugoenhttps://flagos-ai.github.io/FlagGems/performance/overview/Mon, 01 Jan 0001 00:00:00 +0000https://flagos-ai.github.io/FlagGems/performance/overview/<!-- Copyright 2026 FlagOS Contributors Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. --> <h1 id="performance-benchmarking-overview">Performance Benchmarking Overview<a class="anchor" href="#performance-benchmarking-overview">#</a></h1> <p><em>FlagGems</em> operators in general provides better or at least comparable performance when compared to operators from the native PyTorch library. We use the <code>triton.testing.do_bench</code> from the Triton project for benchmarking. The kernel data obtained are shown in the following graph.</p>https://flagos-ai.github.io/FlagGems/performance/benchmark/Mon, 01 Jan 0001 00:00:00 +0000https://flagos-ai.github.io/FlagGems/performance/benchmark/<!-- Copyright 2026 FlagOS Contributors Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. --> <h1 id="performance-benchmarking-in-flaggems">Performance Benchmarking in FlagGems<a class="anchor" href="#performance-benchmarking-in-flaggems">#</a></h1> <p>It is recommended to follow the steps below to add test cases for a new operator. These steps apply to Python-based operators as well as C++-wrapped operators.</p>https://flagos-ai.github.io/FlagGems/performance/results/Mon, 01 Jan 0001 00:00:00 +0000https://flagos-ai.github.io/FlagGems/performance/results/<!-- Copyright 2026 FlagOS Contributors Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. --> <h1 id="benchmark-results">Benchmark Results<a class="anchor" href="#benchmark-results">#</a></h1> <p>The following tables show accuracy and performance benchmark results across platforms. Use the tabs to switch between platforms. Columns are sortable and filterable.</p>https://flagos-ai.github.io/FlagGems/performance/database/Mon, 01 Jan 0001 00:00:00 +0000https://flagos-ai.github.io/FlagGems/performance/database/<!-- Copyright 2026 FlagOS Contributors Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. --> <h1 id="performance-datadase-backends">Performance Datadase Backends<a class="anchor" href="#performance-datadase-backends">#</a></h1> <p><em>FlagGems</em> implements a <code>LibCache</code> class for persisting performance benchmark data into a database. The <code>LibCache</code> interacts with the database backend through <code>sqlalchemy</code>, a generic database abstraction library.</p>