TensorRT-LLMs/cpp/tests/unit_tests/executor/executorConfigTest.cpp
2025-03-11 21:13:42 +08:00

119 lines
4.1 KiB
C++

/*
* SPDX-FileCopyrightText: Copyright (c) 2023-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-License-Identifier: NVIDIA TensorRT Source Code License Agreement
*
* NVIDIA CORPORATION, its affiliates and licensors retain all intellectual
* property and proprietary rights in and to this material, related
* documentation and any modifications thereto. Any use, reproduction,
* disclosure or distribution of this material and related documentation
* without an express license agreement from NVIDIA CORPORATION or
* its affiliates is strictly prohibited.
*/
#include "tensorrt_llm/common/tllmException.h"
#include "tensorrt_llm/executor/executor.h"
#include "tensorrt_llm/executor/types.h"
#include <gmock/gmock.h>
#include <gtest/gtest.h>
using ::testing::_;
using ::testing::Invoke;
using namespace tensorrt_llm::executor;
using namespace tensorrt_llm::common;
TEST(ExecutorConfigTest, ctorValidInputs)
{
SchedulerConfig schedulerConfig;
KvCacheConfig kvCacheConfig;
{
auto executorConfig = ExecutorConfig(1);
}
{
auto executorConfig = ExecutorConfig(1, schedulerConfig, kvCacheConfig, true, true, 0);
}
{
auto executorConfig = ExecutorConfig(1, schedulerConfig, kvCacheConfig, true, true, 1000);
}
}
void testInvalid(SizeType32 maxBeamWidth = 1, SchedulerConfig schedulerConfig = SchedulerConfig(),
KvCacheConfig kvCacheConfig = KvCacheConfig(), bool enableChunkedContext = false, bool normalizeLogProbs = true,
SizeType32 iterStatsMaxIterations = 1000, BatchingType batchingType = BatchingType::kINFLIGHT,
std::optional<ParallelConfig> parallelConfig = std::nullopt)
{
try
{
auto executorConfig = ExecutorConfig(maxBeamWidth, schedulerConfig, kvCacheConfig, enableChunkedContext,
normalizeLogProbs, iterStatsMaxIterations);
FAIL() << "Expected TllmException";
}
catch (TllmException& e)
{
EXPECT_THAT(e.what(), testing::HasSubstr("Assertion failed"));
}
catch (std::exception const& e)
{
FAIL() << "Expected TllmException";
}
}
TEST(ExecutorConfigTest, ctorInvalidInputs)
{
testInvalid(0);
testInvalid(-1);
SchedulerConfig schedulerConfig;
KvCacheConfig kvCacheConfig;
// Empty device ids
try
{
ParallelConfig parallelConfigInvalid;
parallelConfigInvalid.setDeviceIds({});
FAIL() << "Expected TllmException";
}
catch (TllmException& e)
{
EXPECT_THAT(e.what(), testing::HasSubstr("Assertion failed"));
}
catch (std::exception const& e)
{
FAIL() << "Expected TllmException";
}
// iter stats negative
ParallelConfig parallelConfigValid;
testInvalid(1, schedulerConfig, kvCacheConfig, true, true, -1, BatchingType::kINFLIGHT, parallelConfigValid);
}
TEST(ExecutorConfigTest, extendedRuntimePerfKnobConfigTest)
{
ExtendedRuntimePerfKnobConfig extendedRuntimePerfKnobConfig;
{
auto executorConfig = ExecutorConfig(1);
executorConfig.setExtendedRuntimePerfKnobConfig(extendedRuntimePerfKnobConfig);
}
{
auto executorConfig = ExecutorConfig(1);
extendedRuntimePerfKnobConfig.setMultiBlockMode(true);
extendedRuntimePerfKnobConfig.setEnableContextFMHAFP32Acc(true);
executorConfig.setExtendedRuntimePerfKnobConfig(extendedRuntimePerfKnobConfig);
}
{
auto executorConfig = ExecutorConfig(1);
extendedRuntimePerfKnobConfig.setMultiBlockMode(true);
extendedRuntimePerfKnobConfig.setMultiBlockMode(false);
extendedRuntimePerfKnobConfig.setEnableContextFMHAFP32Acc(true);
extendedRuntimePerfKnobConfig.setEnableContextFMHAFP32Acc(false);
executorConfig.setExtendedRuntimePerfKnobConfig(extendedRuntimePerfKnobConfig);
}
{
ExtendedRuntimePerfKnobConfig newExtendedRuntimePerfKnobConfig(false, false);
auto executorConfig = ExecutorConfig(1);
newExtendedRuntimePerfKnobConfig.setMultiBlockMode(true);
newExtendedRuntimePerfKnobConfig.setEnableContextFMHAFP32Acc(true);
executorConfig.setExtendedRuntimePerfKnobConfig(newExtendedRuntimePerfKnobConfig);
}
}