mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-01-14 06:27:45 +08:00
16d2467ea8
* Update TensorRT-LLM --------- Co-authored-by: Denis Kayshev <topenkoff@gmail.com> Co-authored-by: akhoroshev <arthoroshev@gmail.com> Co-authored-by: Patrick Reiter Horn <patrick.horn@gmail.com> Update
92 lines
2.5 KiB
C++
92 lines
2.5 KiB
C++
/*
|
|
* SPDX-FileCopyrightText: Copyright (c) 2022-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
|
|
* SPDX-License-Identifier: Apache-2.0
|
|
*
|
|
* 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 <cstdint>
|
|
#include <optional>
|
|
#include <string>
|
|
|
|
namespace tensorrt_llm::common
|
|
{
|
|
// Useful when you want to inject some debug code controllable with env var.
|
|
std::optional<int32_t> getIntEnv(char const* name);
|
|
|
|
std::optional<size_t> getUInt64Env(char const* name);
|
|
|
|
// XQA kernels (optimized kernels for generation phase).
|
|
bool forceXQAKernels();
|
|
|
|
// Whether XQA JIT is enabled.
|
|
//
|
|
// Returns the value of TRTLLM_ENABLE_XQA_JIT env var. If such env var doesn't exist, std::nullopt is returned.
|
|
std::optional<bool> getEnvEnableXQAJIT();
|
|
|
|
// 0 means to use heuristics.
|
|
std::optional<int32_t> getEnvXqaBlocksPerSequence();
|
|
|
|
// Tune the number of blocks per sequence for accuracy/performance purpose.
|
|
bool getEnvMmhaMultiblockDebug();
|
|
|
|
int getEnvMmhaBlocksPerSequence();
|
|
|
|
int getEnvMmhaKernelBlockSize();
|
|
|
|
// Whether PDL is enabled.
|
|
bool getEnvEnablePDL();
|
|
|
|
bool getEnvUseUCXKvCache();
|
|
|
|
bool getEnvUseMPIKvCache();
|
|
|
|
std::string getEnvUCXInterface();
|
|
|
|
bool getEnvDisaggLayerwise();
|
|
|
|
bool getEnvParallelCacheSend();
|
|
|
|
bool getEnvRequestKVCacheSerial();
|
|
|
|
bool getEnvDisableKVCacheTransferOverlap();
|
|
|
|
bool getEnvDisableReceiveKVCacheParallel();
|
|
|
|
bool getEnvTryZCopyForKVCacheTransfer();
|
|
|
|
// Force deterministic behavior for all kernels.
|
|
bool getEnvForceDeterministic();
|
|
|
|
// Force deterministic behavior for MoE plugin.
|
|
bool getEnvForceDeterministicMOE();
|
|
|
|
// Force deterministic behavior for attention plugin.
|
|
bool getEnvForceDeterministicAttention();
|
|
|
|
// Force deterministic behavior for all reduce plugin.
|
|
bool getEnvForceDeterministicAllReduce();
|
|
|
|
// Return the workspace size for custom all reduce kernels.
|
|
// This only works when force deterministic is enabled.
|
|
size_t getEnvAllReduceWorkspaceSize();
|
|
|
|
bool getEnvKVCacheTransferUseAsyncBuffer();
|
|
|
|
size_t getEnvKVCacheSendMaxConcurrenceNum();
|
|
|
|
size_t getEnvMemSizeForKVCacheTransferBuffer();
|
|
|
|
} // namespace tensorrt_llm::common
|