/* * Copyright (c) 2022-2024, NVIDIA CORPORATION. All rights reserved. * * 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. */ #pragma once #include "tensorrt_llm/runtime/common.h" #include #include namespace tensorrt_llm::runtime { class SamplingConfig { using FloatType = float; template using OptVec = std::optional>; public: explicit SamplingConfig(SizeType beamWidth = 1) : beamWidth{beamWidth} { } SizeType beamWidth; OptVec temperature; // [1] or [batch_size] on cpu OptVec minLength; // [1] or [batch_size] on cpu OptVec repetitionPenalty; // [1] or [batch_size] on cpu OptVec presencePenalty; // [1] or [batch_size] on cpu OptVec frequencyPenalty; // [1] or [batch_size] on cpu // sampling layers OptVec topK; // [1] or [batch_size] on cpu OptVec topP; // [1] or [batch_size] on cpu OptVec randomSeed; // [1] or [batch_size] on cpu OptVec topPDecay; // [batch_size], must between [0, 1] OptVec topPMin; // [batch_size], must between [0, 1] OptVec topPResetIds; // [batch_size] // beam search layer OptVec beamSearchDiversityRate; OptVec lengthPenalty; // speculative decoding OptVec draftAcceptanceThreshold; std::optional normalizeLogProbs; }; } // namespace tensorrt_llm::runtime