/*
* Copyright (c) Huawei Technologies Co., Ltd. 2024. 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.
*/
#include <iostream>
#include <string>
#include <functional>
#include <nlohmann/json.hpp>
#include "atb/infer_op_params.h"
#include "atb/train_op_params.h"
#include "atb/operation.h"
#include "operation_factory.h"
using CreateOperationFuncPtr = std::function<atb::Operation *(const nlohmann::json &)>;
static atb::Operation *ActivationOperationCreate(const nlohmann::json ¶mJson)
{
atb::infer::ActivationParam param;
if (paramJson.contains("activationType")) {
param.activationType = atb::infer::ActivationType(paramJson["activationType"].get<int32_t>());
}
if (paramJson.contains("scale")) {
param.scale = paramJson["scale"].get<float>();
}
if (paramJson.contains("dim")) {
param.dim = paramJson["dim"].get<int32_t>();
}
if (paramJson.contains("geluMode")) {
param.geluMode = atb::infer::ActivationParam::GeLUMode(paramJson["geluMode"].get<int>());
}
atb::Operation *op;
CreateOperation(param, &op);
return op;
}
static atb::Operation *AllGatherOperationCreate(const nlohmann::json ¶mJson)
{
atb::infer::AllGatherParam param;
param.rank = paramJson["rank"].get<int>();
param.rankSize = paramJson["rankSize"].get<int>();
if (paramJson.find("rankRoot") != paramJson.end()) {
param.rankRoot = paramJson["rankRoot"].get<int>();
}
if (paramJson.find("backend") != paramJson.end()) {
param.backend = paramJson["backend"].get<std::string>();
}
if (paramJson.contains("commMode")) {
param.commMode = atb::infer::CommMode(paramJson["commMode"].get<int>());
}
if (paramJson.find("rankTableFile") != paramJson.end()) {
param.rankTableFile = paramJson["rankTableFile"].get<std::string>();
}
if (paramJson.find("commDomain") != paramJson.end()) {
param.commDomain = paramJson["commDomain"].get<std::string>();
}
atb::Operation *op;
CreateOperation(param, &op);
return op;
}
static atb::Operation *AllReduceOperationCreate(const nlohmann::json ¶mJson)
{
atb::infer::AllReduceParam param;
param.rank = paramJson["rank"].get<int>();
param.rankSize = paramJson["rankSize"].get<int>();
if (paramJson.find("rankRoot") != paramJson.end()) {
param.rankRoot = paramJson["rankRoot"].get<int>();
}
if (paramJson.find("backend") != paramJson.end()) {
param.backend = paramJson["backend"].get<std::string>();
}
if (paramJson.find("allReduceType") != paramJson.end()) {
param.allReduceType = paramJson["allReduceType"].get<std::string>();
}
if (paramJson.contains("commMode")) {
param.commMode = atb::infer::CommMode(paramJson["commMode"].get<int>());
}
if (paramJson.find("rankTableFile") != paramJson.end()) {
param.rankTableFile = paramJson["rankTableFile"].get<std::string>();
}
if (paramJson.find("commDomain") != paramJson.end()) {
param.commDomain = paramJson["commDomain"].get<std::string>();
}
atb::Operation *op;
CreateOperation(param, &op);
return op;
}
static atb::Operation *AsStridedOperationCreate(const nlohmann::json ¶mJson)
{
atb::infer::AsStridedParam param;
for (auto item : paramJson["size"]) {
param.size.push_back(item.get<int64_t>());
}
for (auto item : paramJson["stride"]) {
param.stride.push_back(item.get<int64_t>());
}
for (auto item : paramJson["offset"]) {
param.offset.push_back(item.get<int64_t>());
}
atb::Operation *op;
CreateOperation(param, &op);
return op;
}
static atb::Operation *BroadcastOperationCreate(const nlohmann::json ¶mJson)
{
atb::infer::BroadcastParam param;
param.rank = paramJson["rank"].get<int>();
param.rankSize = paramJson["rankSize"].get<int>();
if (paramJson.find("rankRoot") != paramJson.end()) {
param.rankRoot = paramJson["rankRoot"].get<int>();
}
if (paramJson.contains("commMode")) {
param.commMode = atb::infer::CommMode(paramJson["commMode"].get<int>());
}
if (paramJson.find("backend") != paramJson.end()) {
param.backend = paramJson["backend"].get<std::string>();
}
if (paramJson.find("rankTableFile") != paramJson.end()) {
param.rankTableFile = paramJson["rankTableFile"].get<std::string>();
}
if (paramJson.find("commDomain") != paramJson.end()) {
param.commDomain = paramJson["commDomain"].get<std::string>();
}
atb::Operation *op;
CreateOperation(param, &op);
return op;
}
static atb::Operation *ConcatOperationCreate(const nlohmann::json ¶mJson)
{
atb::infer::ConcatParam param;
if (paramJson.contains("concatDim")) {
param.concatDim = paramJson["concatDim"].get<int>();
}
atb::Operation *op;
CreateOperation(param, &op);
return op;
}
static atb::Operation *CumsumOperationCreate(const nlohmann::json ¶mJson)
{
atb::infer::CumsumParam param;
for (auto item : paramJson["axes"]) {
param.axes.push_back(item.get<int64_t>());
}
if (paramJson.contains("exclusive")) {
param.exclusive = paramJson["exclusive"].get<bool>();
}
if (paramJson.contains("reverse")) {
param.reverse = paramJson["reverse"].get<bool>();
}
atb::Operation *op;
CreateOperation(param, &op);
return op;
}
static atb::Operation *ElewiseOperationCreate(const nlohmann::json ¶mJson)
{
atb::infer::ElewiseParam param;
param.elewiseType = paramJson["elewiseType"].get<atb::infer::ElewiseParam::ElewiseType>();
if (paramJson.contains("varAttr")) {
param.mulsParam.varAttr = paramJson["varAttr"].get<float>();
}
if (paramJson.contains("outTensorType")) {
param.outTensorType = paramJson["outTensorType"].get<aclDataType>();
}
if (paramJson.contains("inputScale")) {
param.quantParam.inputScale = paramJson["inputScale"].get<float>();
}
if (paramJson.contains("inputOffset")) {
param.quantParam.inputOffset = paramJson["inputOffset"].get<int>();
}
if (paramJson.contains("asymmetric")) {
param.quantParam.asymmetric = paramJson["asymmetric"].get<bool>();
}
atb::Operation *op;
CreateOperation(param, &op);
return op;
}
static atb::Operation *FastSoftMaxGradOperationCreate(const nlohmann::json ¶mJson)
{
atb::train::FastSoftMaxGradParam param;
if (paramJson.contains("headNum")) {
param.headNum = paramJson["headNum"].get<int32_t>();
}
if (paramJson.contains("qSeqLen")) {
for (auto item : paramJson["qSeqLen"]) {
param.qSeqLen.push_back(item.get<int32_t>());
}
}
atb::Operation *op;
CreateOperation(param, &op);
return op;
}
static atb::Operation *FastSoftMaxOperationCreate(const nlohmann::json ¶mJson)
{
atb::train::FastSoftMaxParam param;
if (paramJson.contains("headNum")) {
param.headNum = paramJson["headNum"].get<int32_t>();
}
if (paramJson.contains("qSeqLen")) {
for (auto item : paramJson["qSeqLen"]) {
param.qSeqLen.push_back(item.get<int32_t>());
}
}
atb::Operation *op;
CreateOperation(param, &op);
return op;
}
static atb::Operation *FillOperationCreate(const nlohmann::json ¶mJson)
{
atb::infer::FillParam param;
if (paramJson.contains("withMask")) {
param.withMask = pa
基于Python的AI一体化开发与调试工具设计源码 （856个子文件） 
install.bat 3KB model.cfg 692B aipp.config 320B operation_creator.cpp 34KB main.cpp 5KB graph_utils.cpp 3KB aie_convert.cpp 2KB export_om_model.cpp 1KB 公网URL使用说明.csv 14KB Dockerfile 6KB
graph_refactor_demo.gif 249KB .gitignore 2KB .gitignore 203B .gitkeep 0B .gitkeep 0B .gitkeep 0B .gitkeep 0B .gitkeep 0B .gitkeep 0B .gitkeep 0B .gitkeep 0B .gitkeep 0B operation_factory.h 1KB graph_utils.h 944B config.ini 668B config.ini 52B
cat.jpg 76KB dog.jpg 13KB test.om.json 509KB model.json 209KB Bloom7BFlashAttentionModel.json 22KB opp.json 2KB model_tree.json 2KB test.onnx.json 444B .keep 0B .keep 0B .keep 0B LICENSE 11KB knowledge_optimizer_rules.md 23KB API_GUIDE.md 21KB 大模型精度问题定位全流程.md 20KB llm大模型推理精度工具功能说明_v0.2.1.md 16KB 大模型推理精度工具说明文档.md 16KB 工具-llm模型迁移分析使用说明.md 15KB graph_refactor_API.md 15KB 工具-精度预检使用说明.md 12KB README_EN.md 12KB 工具-DUMP加速库数据使用说明.md 12KB knowledge_optimizer_framework.md 11KB TorchAir场景-整网算子精度比对.md 9KB 精度预检能力使用说明.md 8KB basic_usage.md 8KB 自动映射比对能力说明.md 7KB 加速库场景-整网精度比对.md 7KB FAQ.md 7KB llm大模型推理精度工具功能说明_v0.2.0.md 6KB 工具-BadCase分析使用说明.md 6KB 工具-DUMP在线推理数据使用说明.md 6KB graph_refactor_BaseGraph.md 6KB introduction.md 5KB 如何识别 Bad Case.md 5KB history.md 5KB graph_refactor_BaseNode.md 5KB 加速库场景-输出Token的logits精度比对.md 5KB FAQ.md 5KB 精度比对结果参数说明.md 5KB 工具-手动映射比对能力说明.md 4KB doc-guidelines.md 4KB 工具-自动比对功能使用说明.md 4KB llm大模型推理精度工具功能说明_v0.1.0.md 3KB 工具-异常检测使用说明.md 3KB MindIE-Torch场景-整网算子精度对比.md 3KB TorchAir场景Dump案例.md 2KB graph_optimizer.md 2KB FAQ.md 2KB README.md 864B FAQ.md 864B FAQ.md 864B OWNERS 736B
torch_topo.png 236KB acc-workflow.png 218KB atb_topo.png 186KB ait_flow.png 172KB msit-flow.png 104KB api_quick_start.png 88KB cmp_report.png 86KB chatglm6b_cmp_result.png 85KB 7bcb6c78-a839-4dad-bb0a-4abcba10694c.png 81KB 比对报告.PNG 69KB msit-llm-flow.png 66KB problem_node.png 66KB bloom7b_cmp_result.png 65KB excerpt.png 63KB basenode.png 62KB LocationProgress.png 49KB inference.png 37KB matched_pie.png 34KB single_node.png 30KB acl_pta_workflow.png 30KB 说明.png 29KB共 856 条
- 1
- 2
- 3
- 4
- 5
- 6
- 9
资源评论
