Python 3.10 & Acuity 6.33 迁移
This commit is contained in:
xujiao
parent
4a112b1f3b
commit
474265b45a
|
|
@ -0,0 +1,104 @@
|
|||
# Netrans Makefile
|
||||
# 统一任务入口，简化开发流程
|
||||
|
||||
.PHONY: help install test clean build package lint format
|
||||
|
||||
# 默认目标
|
||||
help:
|
||||
@echo "Netrans 开发任务管理"
|
||||
@echo ""
|
||||
@echo "可用目标:"
|
||||
@echo " install 安装开发环境 (pip install -e .)"
|
||||
@echo " test 运行所有测试"
|
||||
@echo " test-unit 运行单元测试"
|
||||
@echo " test-int 运行集成测试"
|
||||
@echo " clean 清理临时文件和构建产物"
|
||||
@echo " build 运行 setup.sh 构建"
|
||||
@echo " package 构建离线安装包"
|
||||
@echo " lint 代码格式检查"
|
||||
@echo " format 代码格式化"
|
||||
@echo " sync 同步源码到 devtools/packing/"
|
||||
@echo " docs 构建文档索引"
|
||||
|
||||
# 安装开发环境
|
||||
install:
|
||||
bash setup.sh
|
||||
|
||||
# 运行所有测试
|
||||
test: test-unit test-int
|
||||
@echo "✓ 所有测试完成"
|
||||
|
||||
# 单元测试
|
||||
test-unit:
|
||||
python -m pytest tests/unit/ -v --tb=short 2>/dev/null || \
|
||||
python -m pytest tests/unit/ -v --tb=short || \
|
||||
echo "⚠️ pytest 未安装，跳过单元测试"
|
||||
|
||||
# 集成测试
|
||||
test-int:
|
||||
bash tests/integration/integration_test.sh 2>/dev/null || \
|
||||
echo "⚠️ 集成测试脚本执行失败"
|
||||
|
||||
# 清理临时文件
|
||||
clean:
|
||||
@echo "清理临时文件..."
|
||||
@find . -type d -name "__pycache__" -exec rm -rf {} + 2>/dev/null || true
|
||||
@find . -type f -name "*.pyc" -delete 2>/dev/null || true
|
||||
@find . -type f -name "*.pyo" -delete 2>/dev/null || true
|
||||
@find . -type f -name "*~" -delete 2>/dev/null || true
|
||||
@find . -type f -name ".DS_Store" -delete 2>/dev/null || true
|
||||
@rm -rf build/ dist/ *.egg-info .pytest_cache/ 2>/dev/null || true
|
||||
@echo "✓ 清理完成"
|
||||
|
||||
# 配置环境（添加 PATH）
|
||||
setup:
|
||||
bash setup.sh
|
||||
|
||||
# 构建离线包
|
||||
package:
|
||||
cd devtools/packing && bash build.sh
|
||||
|
||||
# 同步源码到packing
|
||||
sync:
|
||||
@rsync -av --delete src/netrans/ devtools/packing/src/netrans/ 2>/dev/null || \
|
||||
cp -r src/netrans/* devtools/packing/src/netrans/
|
||||
@rsync -av --delete script/ devtools/packing/script/ 2>/dev/null || \
|
||||
cp -r script/* devtools/packing/script/
|
||||
@echo "✓ 源码已同步到 devtools/packing/"
|
||||
|
||||
# 代码检查
|
||||
lint:
|
||||
@which flake8 > /dev/null 2>&1 && flake8 src/netrans/ --max-line-length=120 || \
|
||||
echo "⚠️ flake8 未安装，跳过代码检查"
|
||||
|
||||
# 代码格式化
|
||||
format:
|
||||
@which black > /dev/null 2>&1 && black src/netrans/ || \
|
||||
echo "⚠️ black 未安装，跳过格式化"
|
||||
|
||||
# 构建文档索引
|
||||
docs:
|
||||
@echo "Netrans 文档索引"
|
||||
@echo "================"
|
||||
@echo ""
|
||||
@echo "用户文档:"
|
||||
@echo " docs/quick-start.md - 快速入门"
|
||||
@echo " docs/cli-reference.md - 命令行参考"
|
||||
@echo " docs/netrans_api.md - Python API 参考"
|
||||
@echo ""
|
||||
@echo "功能文档:"
|
||||
@echo " docs/DUMP_USAGE.md - 张量导出说明"
|
||||
@echo " docs/INFERENCE_USAGE.md - 推理功能说明"
|
||||
@echo " docs/FEATURES_UPDATE.md - 功能更新记录"
|
||||
@echo ""
|
||||
@echo "开发文档:"
|
||||
@echo " devtools/ - 开发工具和记录"
|
||||
@echo " tests/ - 测试代码"
|
||||
|
||||
# 检查环境
|
||||
check:
|
||||
@echo "环境检查:"
|
||||
@python3 --version
|
||||
@pip --version
|
||||
@git --version
|
||||
@echo "✓ 基础环境正常"
|
||||
193
README.md
193
README.md
|
|
@ -1,125 +1,142 @@
|
|||
# Netrans 简介
|
||||
|
||||
Netrans 是 Pnna NPU 配套的AI编译器，提供命令行工具 Netrans_cli 和 python API, 其功能是将模型权重转换成在 Pnna NPU 上运行的 nbg（network binary graph）格式文件（.nb 后缀）。 Nbg 文件用于后续模型部署和推理工程的交叉编译。
|
||||
Netrans 是 PNNA NPU 配套的 AI 编译器，提供命令行工具和 Python API，将模型权重转换成在 PNNA NPU 上运行的 NBG（Network Binary Graph）格式文件（.nb 后缀）。NBG 文件用于后续模型部署和推理工程的交叉编译。
|
||||
|
||||
## 工程结构
|
||||
|
||||
Netrans 目录结构如下：
|
||||
|
||||
```text
|
||||
netrans/
|
||||
├── bin # binary file
|
||||
├── docs # 文档，包括用户指南和命令行工具的详细说明
|
||||
├── examples # 示例代码，展示不同框架如何使用netrans进行模型转换
|
||||
├── README.md # 说明文档，通常包含项目概述、安装指南等
|
||||
├── script # 命令行工具
|
||||
├── setup.sh # 用于设置环境或安装依赖的Shell脚本
|
||||
└── test # 测试代码
|
||||
netrans/
|
||||
├── src/netrans # Python 源码
|
||||
├── script # 命令行工具
|
||||
├── docs # 文档
|
||||
├── examples # 示例代码
|
||||
├── vendor # 第三方依赖（acuity whl）
|
||||
├── devtools/packing # 离线打包配置
|
||||
├── devtools # 开发工具（调试脚本等，不提交）
|
||||
├── tests # 单元测试
|
||||
├── test # 集成测试
|
||||
├── setup.py # Python 包配置
|
||||
├── setup.sh # 安装脚本（一键安装）
|
||||
├── requirements.txt # Python 依赖
|
||||
└── README.md # 本文件
|
||||
```
|
||||
|
||||
## 安装指南
|
||||
|
||||
### 系统依赖
|
||||
|
||||
- CPU ： Intel® Core™ i5-6500 CPU @ 3.2 GHz x4 支持 the Intel® Advanced Vector Extensions.
|
||||
- RAM ： 至少8GB
|
||||
- 硬盘 ： 160GB
|
||||
- 操作系统 ： Ubuntu 20.04 LTS 64-bit with Python 3.10，不推荐使用其他版本
|
||||
- **CPU**: Intel® Core™ i5-6500 CPU @ 3.2 GHz x4 或同等性能
|
||||
- **RAM**: 至少 8GB
|
||||
- **硬盘**: 160GB 剩余空间
|
||||
- **操作系统**: Ubuntu 20.04 LTS 64-bit with Python 3.10
|
||||
|
||||
### 安装步骤
|
||||
|
||||
- 安装依赖
|
||||
|
||||
```shell
|
||||
sudo apt update
|
||||
sudo apt install build-essential
|
||||
|
||||
# 安装 mamba ，本项目使用 mamba 创建虚拟环境，演示安装过程
|
||||
|
||||
# 下载 mamba 安装脚本
|
||||
wget "https://mirrors.tuna.tsinghua.edu.cn/github-release/conda-forge/miniforge/LatestRelease/Miniforge3-$(uname)-$(uname -m).sh"
|
||||
# 创建 mamba 的安装目录
|
||||
mkdir -p ~/app
|
||||
# 安装 mamba 到 ~/app/
|
||||
bash Miniforge3-Linux-x86_64.sh -b -p ${HOME}/app/miniforge3
|
||||
# 添加 mamba 的初始化脚本到环境配置文件
|
||||
echo "export MAMBA_ROOT_PREFIX=${HOME}/app/miniforge3" >> ${HOME}/.bashrc
|
||||
echo "source " ${HOME}/app/miniforge3/etc/profile.d/mamba.sh"" >> ${HOME}/.bashrc
|
||||
echo "source " ${HOME}/app/miniforge3/etc/profile.d/conda.sh"" >> ${HOME}/.bashrc
|
||||
# 重新加载 ~/.bashrc 文件，使 mamba 初始化生效
|
||||
source ${HOME}/.bashrc
|
||||
# 创建一个名为 netrans 的虚拟环境，并安装 Python 3.10
|
||||
mamba create -n netrans python=3.10 -y
|
||||
# 激活 netrans 虚拟环境
|
||||
mamba activate netrans
|
||||
```
|
||||
|
||||
- 下载 Netrans
|
||||
|
||||
```bash
|
||||
# 下载 Netrans 到 ～/app
|
||||
cd ~/app
|
||||
git clone https://gitlink.org.cn/nudt_dsp/netrans.git
|
||||
```
|
||||
|
||||
- 运行配置脚本
|
||||
#### 1. 克隆仓库
|
||||
|
||||
```bash
|
||||
git clone https://gitlink.org.cn/nudt_dsp/netrans.git ~/app/netrans
|
||||
cd ~/app/netrans
|
||||
bash setup.sh
|
||||
```
|
||||
|
||||
## Netrans 使用说明
|
||||
#### 2. 创建并激活 Python 3.10 环境
|
||||
|
||||
Netrans 提供 Tensorflow、Caffe、Darknet、ONNX 和 Pytorch 的模型转换示例，请参考目录 `~/app/netrans/examples`
|
||||
```bash
|
||||
# 使用 mamba（推荐）
|
||||
conda install mamba -n base -c conda-forge
|
||||
mamba create -n netrans python=3.10 -y
|
||||
mamba activate netrans
|
||||
|
||||
# 或使用 conda
|
||||
# conda create -n netrans python=3.10 -y
|
||||
# conda activate netrans
|
||||
```
|
||||
|
||||
#### 3. 一键安装
|
||||
|
||||
```bash
|
||||
bash setup.sh
|
||||
source ~/.bashrc
|
||||
```
|
||||
|
||||
`setup.sh` 会依次完成：
|
||||
1. 安装 acuity（vendor/ 目录中的 whl）
|
||||
2. 安装 Python 依赖（requirements.txt）
|
||||
3. 安装 netrans（pip install -e .）
|
||||
4. 添加 script 目录到 PATH
|
||||
|
||||
#### 4. 验证安装
|
||||
|
||||
```bash
|
||||
# 验证 Python 包
|
||||
python -c "import netrans; print(netrans.__version__)"
|
||||
|
||||
# 验证命令行工具
|
||||
netrans --help
|
||||
netrans-dump --help
|
||||
```
|
||||
|
||||
## 快速开始
|
||||
|
||||
### 命令行工具
|
||||
|
||||
Netrans 提供了简单的命令行接口，用于编译和优化模型。
|
||||
|
||||
```bash
|
||||
# 以 转成 ONNX 格式的 YOLOv8s 模型为例，演示使用 Netrans 命令行工具完成转换的全过程。
|
||||
# 1. 定义模型路径，模型路径默认为工作路径。
|
||||
work_path='~/app/netrans/examples/infer_with_pre_post_process/yolov8s'
|
||||
cd ${work_path}
|
||||
# 2. 激活环境
|
||||
# 进入示例目录
|
||||
cd ~/app/netrans/examples/infer_with_pre_post_process/yolov8s
|
||||
|
||||
# 激活环境
|
||||
mamba activate netrans
|
||||
# 3. 模型导入
|
||||
|
||||
# 完整转换流程
|
||||
netrans load ./ --mean 0 0 0 --scale 255 255 255
|
||||
# 4. 模型量化
|
||||
netrans quantize ./ asymu8
|
||||
# 5. 将前后处理加入推理网络
|
||||
netrans quantize ./ asymu8
|
||||
netrans add_pre_post ./ asymu8
|
||||
# 6. 导出 nbg 文件
|
||||
netrans export ./ asymu8
|
||||
netrans export ./ asymu8
|
||||
```
|
||||
|
||||
详细说明请参考[netrans 命令行使用说明](docs/netrans_cli.md)。
|
||||
### Python API
|
||||
|
||||
### Python接口
|
||||
|
||||
通过Netrans Python接口，可以方便地在Python脚本中调用编译器。
|
||||
示例代码：
|
||||
|
||||
```py3
|
||||
```python
|
||||
from netrans import Netrans
|
||||
|
||||
# 定义模型路径，模型路径默认为工作路径。
|
||||
model_path='~/app/netrans/examples/infer_with_pre_post_process/yolov8s'
|
||||
import sys
|
||||
model_path=sys.argv[1]
|
||||
|
||||
# 初始化netrans
|
||||
# 初始化
|
||||
net = Netrans()
|
||||
# 模型载入,同时配置 mean 和 scale
|
||||
net.load(model_path, mean=[0,0,0] ,scale=[255] )
|
||||
# 模型量化
|
||||
net.quantize("asymu8", preprocess = False, postprocess= False)
|
||||
# 前后处理添加进推理
|
||||
net.add_pre_post("asymu8")
|
||||
# 模型导出
|
||||
net.export("asymu8")
|
||||
|
||||
# 加载模型
|
||||
net.load('./yolov8s', mean=[0, 0, 0], scale=[255])
|
||||
|
||||
# 量化
|
||||
net.quantize('asymu8')
|
||||
|
||||
# 添加前后处理
|
||||
net.add_pre_post('asymu8')
|
||||
|
||||
# 导出
|
||||
net.export('asymu8')
|
||||
```
|
||||
|
||||
详细说明请参考[netrans api 使用说明](docs/netrans_py.md)。
|
||||
## 文档
|
||||
|
||||
- [快速入门](docs/quick-start.md)
|
||||
- [命令行参考](docs/cli-reference.md)
|
||||
- [Python API 参考](docs/netrans_api.md)
|
||||
- [版本发布记录](docs/release.md)
|
||||
|
||||
## 开发安装
|
||||
|
||||
如需开发调试，可直接使用 Makefile：
|
||||
|
||||
```bash
|
||||
make install # 执行 setup.sh 的完整安装
|
||||
make setup # 仅配置环境变量
|
||||
make clean # 清理临时文件
|
||||
make package # 构建离线安装包
|
||||
```
|
||||
|
||||
## 离线安装
|
||||
|
||||
如需在无网络环境安装，请参考 [devtools/packing/README.md](devtools/packing/README.md)。
|
||||
|
||||
## 许可证
|
||||
|
||||
与 Netrans 主项目相同。
|
||||
|
|
|
|||
|
|
@ -0,0 +1,11 @@
|
|||
scipy
|
||||
tensorflow>=2.3.0,<=2.15.0
|
||||
protobuf<=3.20.3
|
||||
networkx>=1.11
|
||||
onnx>=1.8.0,<=1.14.0
|
||||
onnxoptimizer>=0.2.5,<=0.3.13
|
||||
dill==0.2.8.2
|
||||
ruamel.yaml<0.18.0
|
||||
ply==3.11
|
||||
torch>=1.5.1,<=2.2.2
|
||||
numpy < 1.24
|
||||
|
|
@ -0,0 +1,58 @@
|
|||
# -*- coding: utf-8 -*-
|
||||
import os
|
||||
from pathlib import Path
|
||||
from setuptools import setup, find_packages
|
||||
|
||||
# 基础目录
|
||||
BASE = Path(__file__).parent
|
||||
SRC_DIR = BASE / "src"
|
||||
|
||||
# 收集 netrans 包（不包含 acuitylib）
|
||||
netrans_packages = find_packages(
|
||||
where=str(SRC_DIR),
|
||||
include=["netrans", "netrans.*"]
|
||||
)
|
||||
|
||||
setup(
|
||||
name="netrans",
|
||||
version="6.33.0",
|
||||
description="Netrans inference SDK - Neural Network Transformation Toolkit for PNNA Chips",
|
||||
author="Your Name",
|
||||
author_email="you@example.com",
|
||||
url="https://github.com/yourname/netrans",
|
||||
license="Proprietary",
|
||||
python_requires=">=3.10",
|
||||
|
||||
# 关键：告诉 setuptools 去哪里找包
|
||||
package_dir={
|
||||
"": "src", # 所有包都在 src 下
|
||||
},
|
||||
packages=netrans_packages,
|
||||
include_package_data=True,
|
||||
zip_safe=False,
|
||||
|
||||
# 入口点 - 创建命令行脚本
|
||||
entry_points={
|
||||
'console_scripts': [
|
||||
'netrans-dump=netrans.dump:main',
|
||||
'netrans-export-nbg=netrans.export_nbg:main',
|
||||
'netrans-inference=netrans.inference:main',
|
||||
'netrans-importer=netrans.importer:main',
|
||||
'netrans-measure=netrans.measure:main',
|
||||
'netrans-quantize=netrans.quantize:main',
|
||||
'netrans-quantize-hybrid=netrans.quantize_hybrid:main',
|
||||
'netrans-add-prepost=netrans.add_prepost_to_graph:main',
|
||||
],
|
||||
},
|
||||
|
||||
# 依赖说明（acuity 需要单独安装）
|
||||
extras_require={
|
||||
'acuity': ['acuity>=6.33.0'],
|
||||
},
|
||||
|
||||
classifiers=[
|
||||
"Development Status :: 4 - Beta",
|
||||
"Intended Audience :: Developers",
|
||||
"Programming Language :: Python :: 3.10",
|
||||
],
|
||||
)
|
||||
108
setup.sh
108
setup.sh
|
|
@ -1,65 +1,71 @@
|
|||
#!/bin/bash
|
||||
|
||||
# === 脚本目录 ===
|
||||
# Netrans 安装脚本
|
||||
# 功能：
|
||||
# 1. 安装 acuity（核心依赖）
|
||||
# 2. 安装 Python 依赖
|
||||
# 3. 安装 netrans（可编辑模式）
|
||||
# 4. 添加命令行工具到 PATH
|
||||
|
||||
set -e
|
||||
|
||||
CURRENT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
|
||||
SCRIPT_DIR="$CURRENT_DIR/script"
|
||||
# SRC_DIR="$CURRENT_DIR/src"
|
||||
|
||||
# 定义要追加到 .bashrc 的环境变量
|
||||
ENV_VARS=(
|
||||
"export PATH=\"\$PATH:$SCRIPT_DIR\""
|
||||
# "export PATH=\"\$PATH:$SRC_DIR\""
|
||||
# "export ACUITY_LOG_LEVEL=ERROR"
|
||||
)
|
||||
echo "========================================"
|
||||
echo " Netrans 安装脚本"
|
||||
echo "========================================"
|
||||
|
||||
# 扫描 .bashrc，逐条添加不存在的变量
|
||||
for LINE in "${ENV_VARS[@]}"; do
|
||||
if grep -Fxq "$LINE" ~/.bashrc; then
|
||||
echo "已存在：$LINE"
|
||||
else
|
||||
echo "$LINE" >> ~/.bashrc
|
||||
echo "已添加：$LINE"
|
||||
fi
|
||||
done
|
||||
# 步骤 1: 安装 acuity
|
||||
echo ""
|
||||
echo "[1/4] 安装 acuity..."
|
||||
ACUITY_WHL="$CURRENT_DIR/vendor/acuity-6.33.19-cp310-cp310-manylinux2010_x86_64.whl"
|
||||
if [ -f "$ACUITY_WHL" ]; then
|
||||
pip install "$ACUITY_WHL" --no-deps
|
||||
echo "✓ acuity 安装完成"
|
||||
else
|
||||
echo "✗ 未找到 acuity whl 文件: $ACUITY_WHL"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
pip install bin/netrans*.whl --force-reinstall
|
||||
pip install -r requirements_py3.10.txt
|
||||
# 步骤 2: 安装 Python 依赖
|
||||
echo ""
|
||||
echo "[2/4] 安装 Python 依赖..."
|
||||
pip install -r "$CURRENT_DIR/requirements.txt"
|
||||
echo "✓ 依赖安装完成"
|
||||
|
||||
# # === 处理 TensorFlow XLA JIT 编译问题 ===
|
||||
# echo ""
|
||||
# echo "正在检查 TensorFlow XLA JIT 编译依赖..."
|
||||
# 步骤 3: 安装 netrans
|
||||
echo ""
|
||||
echo "[3/4] 安装 netrans..."
|
||||
pip install -e "$CURRENT_DIR"
|
||||
echo "✓ netrans 安装完成"
|
||||
|
||||
# # 确保 cuda-nvcc 已安装（提供 libdevice.10.bc）
|
||||
# if ! conda list cuda-nvcc &>/dev/null; then
|
||||
# echo "cuda-nvcc 未找到，正在安装 cuda-nvcc=12.3..."
|
||||
# conda install -c nvidia cuda-nvcc=12.3 -y
|
||||
# fi
|
||||
# 步骤 4: 添加 PATH
|
||||
echo ""
|
||||
echo "[4/4] 配置环境变量..."
|
||||
if [ ! -d "$SCRIPT_DIR" ]; then
|
||||
echo "✗ 未找到 script 目录"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
# # 查找 libdevice.10.bc 文件
|
||||
# LIBDEVICE_PATH=$(find $CONDA_PREFIX -name "libdevice.10.bc" 2>/dev/null | head -n 1)
|
||||
# if [[ -n "$LIBDEVICE_PATH" ]]; then
|
||||
# LIBDEVICE_DIR=$(dirname "$LIBDEVICE_PATH")
|
||||
# echo "找到 libdevice.10.bc: $LIBDEVICE_PATH"
|
||||
|
||||
# # 设置 XLA_FLAGS 环境变量
|
||||
# XLA_FLAG="export XLA_FLAGS=--xla_gpu_cuda_data_dir=$LIBDEVICE_DIR"
|
||||
|
||||
# # 检查是否已存在在 .bashrc 中
|
||||
# if grep -Fxq "$XLA_FLAG" ~/.bashrc; then
|
||||
# echo "XLA_FLAGS 已存在：$XLA_FLAG"
|
||||
# else
|
||||
# echo "$XLA_FLAG" >> ~/.bashrc
|
||||
# echo "已添加 XLA_FLAGS 到 .bashrc: $XLA_FLAG"
|
||||
# fi
|
||||
# else
|
||||
# echo "警告: 未找到 libdevice.10.bc，即使已安装 cuda-nvcc"
|
||||
# echo "请手动检查 CUDA 安装或设置 XLA_FLAGS"
|
||||
# fi
|
||||
LINE="export PATH=\"\$PATH:$SCRIPT_DIR\""
|
||||
if grep -Fxq "$LINE" ~/.bashrc; then
|
||||
echo "PATH 已配置"
|
||||
else
|
||||
echo "" >> ~/.bashrc
|
||||
echo "# Netrans 命令行工具路径" >> ~/.bashrc
|
||||
echo "$LINE" >> ~/.bashrc
|
||||
echo "✓ PATH 已添加"
|
||||
fi
|
||||
|
||||
echo ""
|
||||
echo "script path 已添加到 ~/.bashrc"
|
||||
echo "请运行以下命令使其立即生效："
|
||||
echo "========================================"
|
||||
echo " 安装完成！"
|
||||
echo "========================================"
|
||||
echo ""
|
||||
echo " source ~/.bashrc"
|
||||
echo "请执行以下命令使配置生效："
|
||||
echo " source ~/.bashrc"
|
||||
echo ""
|
||||
|
||||
echo "验证安装："
|
||||
echo " python -c \"import netrans; print(netrans.__version__)\""
|
||||
echo " netrans --help"
|
||||
|
|
|
|||
|
|
@ -5,5 +5,5 @@ Netrans - PNNA AI 编译器
|
|||
|
||||
from .netrans import Netrans
|
||||
|
||||
__version__ = "6.42.3"
|
||||
__version__ = "6.33.0"
|
||||
__all__ = ["Netrans"]
|
||||
|
|
@ -36,8 +36,8 @@ FATAL: Core dependency 'acuitylib' is not available.
|
|||
|
||||
Netrans requires acuitylib to function. Please install or configure it:
|
||||
|
||||
1. Install acuitylib:
|
||||
pip install acuitylib
|
||||
1. Install acuitylib (use --no-deps to avoid overwriting torch):
|
||||
pip install acuitylib --no-deps
|
||||
|
||||
2. Or set ACUITY_PATH environment variable:
|
||||
export ACUITY_PATH=/path/to/acuity
|
||||
|
|
@ -55,7 +55,7 @@ FATAL: Core dependency 'acuitylib' is not available and ACUITY_PATH is not set.
|
|||
|
||||
Netrans cannot function without acuitylib. Please:
|
||||
|
||||
1. Install acuitylib: pip install acuitylib
|
||||
1. Install acuitylib: pip install acuitylib --no-deps
|
||||
2. Or set ACUITY_PATH: export ACUITY_PATH=/path/to/acuity
|
||||
|
||||
Original error: {e}
|
||||
|
|
|
|||
|
|
@ -90,6 +90,79 @@ def export_nbg(net, model_filename, quantized, optimize, viv_sdk=None, use_hybri
|
|||
optimize=optimize, viv_sdk=viv_sdk, pack_nbg_unify=True)
|
||||
nn.export_ovxlib(net, output_path=output_dir, dtype=quantized, optimize=optimize, viv_sdk=viv_sdk, pack_nbg_unify=True)
|
||||
|
||||
|
||||
def export_nbg_without_reload(net, model_filename, quantized, optimize, viv_sdk=None, use_hybrid=False):
|
||||
"""
|
||||
直接使用已加载的网络进行导出，避免重新加载导致配置重置。
|
||||
|
||||
此函数不依赖 load_net() 设置的全局变量 post，而是直接构造文件路径，
|
||||
确保 add_pre_post 修改的 inputmeta 和 postprocess 配置不会被重置。
|
||||
|
||||
关键：在导出前保存网络配置，确保 nn.export_ovxlib 重新加载时能拿到最新配置。
|
||||
|
||||
Args:
|
||||
net: 已加载的网络对象（直接使用 self._meta.net）
|
||||
model_filename: 模型文件名前缀
|
||||
quantized: 量化类型
|
||||
optimize: 优化配置
|
||||
viv_sdk: Vivante SDK 路径（可选）
|
||||
use_hybrid: 是否使用 hybrid 量化
|
||||
"""
|
||||
nn = VSInn()
|
||||
if not use_hybrid:
|
||||
quantize_file = model_filename + '_' + quantized + ".quantize"
|
||||
model = model_filename + ".json"
|
||||
output_dir = 'wksp/{}_{}'.format(model_filename, quantized)
|
||||
else:
|
||||
# add hybrid quantize for print log
|
||||
quantize_file = model_filename + '_' + quantized + "_hy.quantize"
|
||||
model = model_filename + '_' + quantized + "_hy.quantize.json"
|
||||
# add hybrid quantize output file name
|
||||
output_dir = 'wksp/{}_{}'.format(model_filename, quantized + "_hy")
|
||||
output_dir = os.path.join(output_dir, os.path.split(output_dir)[1])
|
||||
|
||||
# 直接构造后处理文件路径，不依赖全局变量 post
|
||||
# 这样确保使用当前磁盘上的文件（已被 add_pre_post 修改过）
|
||||
postprocess_file = model_filename + "_postprocess_file.yml"
|
||||
if not os.path.exists(postprocess_file):
|
||||
postprocess_file = None
|
||||
|
||||
# 关键：在导出前，把磁盘上的配置加载到网络对象
|
||||
# 因为 add_pre_post 只修改了 YAML 文件，没有更新 net 对象
|
||||
# nn.export_ovxlib 会从 net 对象读取配置，所以需要先同步
|
||||
try:
|
||||
inputmeta_yml = model_filename + '_inputmeta.yml'
|
||||
if os.path.exists(inputmeta_yml):
|
||||
nn.load_model_inputmeta(net, inputmeta_yml)
|
||||
print(f" 已加载预处理配置: {inputmeta_yml}")
|
||||
if postprocess_file and os.path.exists(postprocess_file):
|
||||
nn.load_model_outputmeta(net, postprocess_file)
|
||||
print(f" 已加载后处理配置: {postprocess_file}")
|
||||
except Exception as e:
|
||||
print(f"Warning: Failed to load model config before export: {e}")
|
||||
|
||||
if quantized != 'float32':
|
||||
print_params(nn.export_ovxlib, model=model, data=model_filename + ".data", quantize=quantize_file,
|
||||
with_input_meta=model_filename + "_inputmeta.yml", postprocess_file=postprocess_file, output_path=output_dir,
|
||||
optimize=optimize, viv_sdk=viv_sdk, pack_nbg_unify=True)
|
||||
else:
|
||||
print_params(nn.export_ovxlib, model=model, data=model_filename + ".data",
|
||||
with_input_meta=model_filename + "_inputmeta.yml", postprocess_file=postprocess_file, output_path=output_dir,
|
||||
optimize=optimize, viv_sdk=viv_sdk, pack_nbg_unify=True)
|
||||
|
||||
# 关键：传入 with_input_meta 和 postprocess_file 参数，确保 export_ovxlib 使用正确的配置文件
|
||||
# 而不是重新生成它们
|
||||
nn.export_ovxlib(
|
||||
net,
|
||||
output_path=output_dir,
|
||||
dtype=quantized,
|
||||
optimize=optimize,
|
||||
viv_sdk=viv_sdk,
|
||||
pack_nbg_unify=True,
|
||||
with_input_meta=model_filename + "_inputmeta.yml",
|
||||
postprocess_file=postprocess_file
|
||||
)
|
||||
|
||||
# generate the execution file cmd.sh and move the tensors generated by infernece.py
|
||||
def generate_exe_script(net, model_filename, quantized, use_hybrid=False):
|
||||
if use_hybrid:
|
||||
|
|
|
|||
|
|
@ -18,6 +18,7 @@ from .config import (
|
|||
_ACUITY_AVAILABLE, ChannelParams
|
||||
)
|
||||
|
||||
|
||||
# Local Module Imports
|
||||
from .utils import get_modelfile_name
|
||||
from .decorators import chdir, validate_quantization_type, validate_platform
|
||||
|
|
@ -258,8 +259,8 @@ class Netrans:
|
|||
"""
|
||||
Load and prepare a model for processing.
|
||||
|
||||
This method imports the model, updates metadata, saves preprocessing
|
||||
parameters, and applies pre/post-processing transformations.
|
||||
This method imports the model from original format (like .cfg, .weights, .pb, etc.),
|
||||
generates .json and .data files, and applies preprocessing/postprocessing.
|
||||
|
||||
Args:
|
||||
model_path: Path to the model directory containing model files
|
||||
|
|
@ -295,7 +296,7 @@ class Netrans:
|
|||
try:
|
||||
os.chdir(abs_path)
|
||||
|
||||
# Import model using standard importer
|
||||
# Import model using standard importer (from original format)
|
||||
_, _ = importer(model_filename)
|
||||
|
||||
# Update metadata - 使用绝对路径避免相对路径问题
|
||||
|
|
@ -311,14 +312,81 @@ class Netrans:
|
|||
finally:
|
||||
os.chdir(original_cwd)
|
||||
|
||||
def load_generated(self, model_path: str, quantized: str = "float32",
|
||||
use_hybrid: bool = False) -> None:
|
||||
"""
|
||||
Load model from generated files (.json, .data, .quantize, _inputmeta.yml).
|
||||
|
||||
This method is used for quantize, export, and other operations that work
|
||||
on already imported models. It loads the model from generated files instead
|
||||
of re-importing from original format.
|
||||
|
||||
Args:
|
||||
model_path: Path to the model directory containing generated files
|
||||
quantized: Quantization type (default: "float32")
|
||||
use_hybrid: Whether to use hybrid quantization files (default: False)
|
||||
|
||||
Raises:
|
||||
FileNotFoundError: If model directory or required files don't exist
|
||||
ValueError: If model files are not found
|
||||
|
||||
Example:
|
||||
>>> model = Netrans()
|
||||
>>> model.load_generated('path/to/model', quantized='asymu8')
|
||||
"""
|
||||
# Validate model path
|
||||
abs_path = os.path.abspath(model_path)
|
||||
if not (os.path.exists(abs_path) and os.path.isdir(abs_path)):
|
||||
raise FileNotFoundError("Please enter the path that includes the model.")
|
||||
|
||||
# Find model filename
|
||||
model_filename = get_modelfile_name(abs_path)
|
||||
if model_filename is None:
|
||||
raise ValueError("Cannot find model file under given path.")
|
||||
|
||||
# Change to model directory temporarily
|
||||
original_cwd = os.getcwd()
|
||||
try:
|
||||
os.chdir(abs_path)
|
||||
|
||||
# Check required files exist
|
||||
json_file = f"{model_filename}.json"
|
||||
data_file = f"{model_filename}.data"
|
||||
inputmeta_file = f"{model_filename}_inputmeta.yml"
|
||||
|
||||
if not os.path.exists(json_file):
|
||||
raise FileNotFoundError(f"Model file not found: {json_file}. "
|
||||
f"Please run 'netrans load' first.")
|
||||
if not os.path.exists(data_file):
|
||||
raise FileNotFoundError(f"Model data file not found: {data_file}. "
|
||||
f"Please run 'netrans load' first.")
|
||||
|
||||
print(f"Loading from generated files: {json_file}, {data_file}")
|
||||
|
||||
# Load model using export_nbg's load_net logic
|
||||
from .export_nbg import load_net as load_net_for_export
|
||||
net = load_net_for_export(model_filename, quantized, use_hybrid)
|
||||
|
||||
# Update metadata
|
||||
self._meta = ModelMeta(
|
||||
path=abs_path,
|
||||
name=model_filename,
|
||||
net=net,
|
||||
use_hybrid=use_hybrid
|
||||
)
|
||||
|
||||
print(f"✓ Model loaded from generated files: {model_filename}")
|
||||
|
||||
finally:
|
||||
os.chdir(original_cwd)
|
||||
|
||||
@_ensure_meta
|
||||
@chdir
|
||||
def quantize(self, quantized: str, *, model_path: Optional[str] = None,
|
||||
algorithm: int = 1, iterations: int = 1, entropy: bool = False,
|
||||
mle: bool = False, lid: Optional[str] = None,
|
||||
in_out_quantized: Optional[str] = None,
|
||||
quantize_file: Optional[str] = None, preprocess: bool = False,
|
||||
postprocess: bool = False) -> None:
|
||||
quantize_file: Optional[str] = None) -> None:
|
||||
"""
|
||||
Quantize the loaded model to specified format.
|
||||
|
||||
|
|
@ -336,8 +404,6 @@ class Netrans:
|
|||
lid: Path to JSON file with input/output layer names
|
||||
in_out_quantized: Path to JSON file with input/output quantization types
|
||||
quantize_file: Path to quantization file (required for QAT)
|
||||
pre: Whether to integrate preprocessing into inference graph
|
||||
post: Whether to integrate postprocessing into inference graph
|
||||
|
||||
Raises:
|
||||
QuantizationError: If quantization fails
|
||||
|
|
@ -358,10 +424,6 @@ class Netrans:
|
|||
lid,
|
||||
in_out_quantized,
|
||||
)
|
||||
|
||||
if preprocess or postprocess:
|
||||
self.add_pre_post(quantized, preprocess=preprocess, postprocess=postprocess)
|
||||
|
||||
except Exception as e:
|
||||
raise QuantizationError(f"Quantization failed: {e}") from e
|
||||
|
||||
|
|
@ -369,12 +431,14 @@ class Netrans:
|
|||
@chdir
|
||||
def export(self, quantized: str = "float32", *, model_path: Optional[str] = None,
|
||||
platform: str = "pnna", viv_sdk: Optional[str] = None,
|
||||
use_hybrid: bool = False) -> None:
|
||||
use_hybrid: bool = False, preprocess: bool = False,
|
||||
postprocess: bool = False) -> None:
|
||||
"""
|
||||
Export the quantized model to chip-specific format.
|
||||
|
||||
Converts the model to NBG (Network Binary Graph) format for deployment
|
||||
on PNNA chips. Supports different chip platforms and configurations.
|
||||
Optionally integrates preprocessing and postprocessing into the inference graph.
|
||||
|
||||
Args:
|
||||
quantized: Quantization type used for export (default: "float32")
|
||||
|
|
@ -384,6 +448,8 @@ class Netrans:
|
|||
- 'pnna2': VIP9400O_PID0X1000004F
|
||||
viv_sdk: Path to Vivante SDK (optional)
|
||||
use_hybrid: Whether to use hybrid quantization files
|
||||
preprocess: Whether to integrate preprocessing into inference graph (default: False)
|
||||
postprocess: Whether to integrate postprocessing into inference graph (default: False)
|
||||
|
||||
Raises:
|
||||
ExportError: If export fails
|
||||
|
|
@ -391,18 +457,25 @@ class Netrans:
|
|||
|
||||
Example:
|
||||
>>> model.export('asymu8', platform='pnna')
|
||||
>>> model.export('asymu8', platform='pnna2')
|
||||
>>> model.export('asymu8', platform='pnna2', preprocess=True, postprocess=True)
|
||||
"""
|
||||
# Validate quantization type and platform
|
||||
validate_quantization_type(quantized, allow_float32=True)
|
||||
optimize = validate_platform(platform)
|
||||
|
||||
try:
|
||||
from .export_nbg import load_net
|
||||
temp_net = load_net(self._meta.name, quantized, use_hybrid)
|
||||
# 1. 如果需要，先修改 pre/post 配置（在导出前，直接使用已加载的网络）
|
||||
if preprocess or postprocess:
|
||||
if 'fp16' not in quantized:
|
||||
self.add_pre_post(quantized,
|
||||
preprocess=preprocess,
|
||||
postprocess=postprocess,
|
||||
use_hybrid=use_hybrid)
|
||||
|
||||
export_nbg_acuity(
|
||||
temp_net,
|
||||
# 2. 直接使用已加载的网络导出，避免重新加载导致配置重置
|
||||
from .export_nbg import export_nbg_without_reload
|
||||
export_nbg_without_reload(
|
||||
self._meta.net,
|
||||
self._meta.name,
|
||||
quantized,
|
||||
optimize,
|
||||
|
|
|
|||
|
|
@ -0,0 +1,233 @@
|
|||
#!/bin/bash
|
||||
# 自动化测试脚本：循环处理examples目录下的模型
|
||||
# 使用方法: ./run_examples_test.sh [模型类型] [测试步骤]
|
||||
# 示例: ./run_examples_test.sh onnx all
|
||||
# ./run_examples_test.sh all 01_load
|
||||
|
||||
set -e # 遇到错误时退出
|
||||
|
||||
# 颜色输出
|
||||
RED='\033[0;31m'
|
||||
GREEN='\033[0;32m'
|
||||
YELLOW='\033[1;33m'
|
||||
NC='\033[0m' # No Color
|
||||
|
||||
# 默认参数
|
||||
MODEL_TYPE=${1:-"onnx"} # 默认测试ONNX模型
|
||||
TEST_STEP=${2:-"all"} # 默认运行所有步骤
|
||||
|
||||
# 基础路径
|
||||
EXAMPLES_DIR="/home/xj/work/nudt/netrans/examples"
|
||||
NETRANS_SCRIPT="/home/xj/work/nudt/netrans/script/netrans"
|
||||
TEST_DIR="/home/xj/work/nudt/netrans/test/netrans_cli"
|
||||
|
||||
# 检查netrans脚本是否存在
|
||||
if [ ! -f "$NETRANS_SCRIPT" ]; then
|
||||
echo -e "${RED}错误: netrans脚本不存在: $NETRANS_SCRIPT${NC}"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
# 检查测试目录是否存在
|
||||
if [ ! -d "$TEST_DIR" ]; then
|
||||
echo -e "${RED}错误: 测试目录不存在: $TEST_DIR${NC}"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
# 获取指定类型的模型列表
|
||||
get_models() {
|
||||
local type=$1
|
||||
case $type in
|
||||
"onnx")
|
||||
find "$EXAMPLES_DIR" -name "*.onnx" -type f | grep -v quantize_hybrid | head -5
|
||||
;;
|
||||
"caffe")
|
||||
find "$EXAMPLES_DIR" -name "*.caffemodel" -type f | head -5
|
||||
;;
|
||||
"tensorflow")
|
||||
find "$EXAMPLES_DIR" -name "*.pb" -type f | head -5
|
||||
;;
|
||||
"paddle")
|
||||
find "$EXAMPLES_DIR" -name "*.pdmodel" -type f | head -5
|
||||
;;
|
||||
"all")
|
||||
find "$EXAMPLES_DIR" -name "*.onnx" -o -name "*.caffemodel" -o -name "*.pb" -o -name "*.pdmodel" | head -10
|
||||
;;
|
||||
*)
|
||||
echo -e "${RED}不支持的模型类型: $type${NC}"
|
||||
echo "支持的类型: onnx, caffe, tensorflow, paddle, all"
|
||||
exit 1
|
||||
;;
|
||||
esac
|
||||
}
|
||||
|
||||
# 获取模型对应的目录
|
||||
get_model_dir() {
|
||||
local model_path=$1
|
||||
dirname "$model_path"
|
||||
}
|
||||
|
||||
# 获取模型名称（不含路径和扩展名）
|
||||
get_model_name() {
|
||||
local model_path=$1
|
||||
basename "$model_path" | sed 's/\..*$//'
|
||||
}
|
||||
|
||||
# 运行指定步骤
|
||||
run_step() {
|
||||
local step_script=$1
|
||||
local work_dir=$2
|
||||
local model_name=$3
|
||||
|
||||
echo -e "${YELLOW}运行 $step_script - 模型: $model_name${NC}"
|
||||
echo -e "${YELLOW}工作目录: $work_dir${NC}"
|
||||
|
||||
local full_script_path="$TEST_DIR/$step_script"
|
||||
if [ -f "$full_script_path" ]; then
|
||||
if bash "$full_script_path" "$work_dir"; then
|
||||
echo -e "${GREEN}✓ $step_script 完成${NC}"
|
||||
return 0
|
||||
else
|
||||
echo -e "${RED}✗ $step_script 失败${NC}"
|
||||
return 1
|
||||
fi
|
||||
else
|
||||
echo -e "${RED}步骤脚本不存在: $full_script_path${NC}"
|
||||
return 1
|
||||
fi
|
||||
}
|
||||
|
||||
# 运行完整测试流程
|
||||
run_full_test() {
|
||||
local work_dir=$1
|
||||
local model_name=$2
|
||||
|
||||
echo -e "${YELLOW}=== 开始完整测试流程: $model_name ===${NC}"
|
||||
|
||||
# 定义步骤顺序
|
||||
local steps=(
|
||||
"01_load.sh"
|
||||
"02_quantize.sh"
|
||||
"03_quantize_hybrid.sh"
|
||||
"04_add_pre_post.sh"
|
||||
"05_export.sh"
|
||||
"06_inference.sh"
|
||||
"07_inference_hybrid.sh"
|
||||
"08_dump.sh"
|
||||
"09_add_pre_post.sh"
|
||||
)
|
||||
|
||||
for step in "${steps[@]}"; do
|
||||
if ! run_step "$step" "$work_dir" "$model_name"; then
|
||||
echo -e "${RED}测试流程在 $step 中断${NC}"
|
||||
return 1
|
||||
fi
|
||||
echo
|
||||
done
|
||||
|
||||
echo -e "${GREEN}=== 完整测试流程完成: $model_name ===${NC}"
|
||||
return 0
|
||||
}
|
||||
|
||||
# 运行指定步骤
|
||||
run_single_step() {
|
||||
local step_name=$1
|
||||
local work_dir=$2
|
||||
local model_name=$3
|
||||
|
||||
local step_script="${step_name}.sh"
|
||||
run_step "$step_script" "$work_dir" "$model_name"
|
||||
}
|
||||
|
||||
# 主函数
|
||||
main() {
|
||||
echo -e "${YELLOW}=== Netrans Examples 自动化测试开始 ===${NC}"
|
||||
echo -e "测试模型类型: $MODEL_TYPE"
|
||||
echo -e "测试步骤: $TEST_STEP"
|
||||
echo
|
||||
|
||||
# 获取模型列表
|
||||
models=$(get_models "$MODEL_TYPE")
|
||||
if [ -z "$models" ]; then
|
||||
echo -e "${RED}未找到任何$MODEL_TYPE类型的模型${NC}"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
echo -e "${GREEN}找到以下模型:${NC}"
|
||||
echo "$models" | nl
|
||||
echo
|
||||
|
||||
# 统计结果
|
||||
local total=0
|
||||
local success=0
|
||||
local failed=0
|
||||
|
||||
# 对每个模型进行测试
|
||||
while IFS= read -r model_path; do
|
||||
if [ -z "$model_path" ]; then
|
||||
continue
|
||||
fi
|
||||
|
||||
total=$((total + 1))
|
||||
local model_dir=$(get_model_dir "$model_path")
|
||||
local model_name=$(get_model_name "$model_path")
|
||||
|
||||
echo -e "${YELLOW}--- 处理第 $total 个模型 ---${NC}"
|
||||
|
||||
if [ "$TEST_STEP" = "all" ]; then
|
||||
if run_full_test "$model_dir" "$model_name"; then
|
||||
success=$((success + 1))
|
||||
else
|
||||
failed=$((failed + 1))
|
||||
fi
|
||||
else
|
||||
if run_single_step "$TEST_STEP" "$model_dir" "$model_name"; then
|
||||
success=$((success + 1))
|
||||
else
|
||||
failed=$((failed + 1))
|
||||
fi
|
||||
fi
|
||||
|
||||
echo
|
||||
echo "----------------------------------------"
|
||||
echo
|
||||
done <<< "$models"
|
||||
|
||||
# 输出统计结果
|
||||
echo -e "${YELLOW}=== 测试统计 ===${NC}"
|
||||
echo -e "总模型数: $total"
|
||||
echo -e "${GREEN}成功: $success${NC}"
|
||||
echo -e "${RED}失败: $failed${NC}"
|
||||
|
||||
if [ $failed -eq 0 ]; then
|
||||
echo -e "${GREEN}所有测试都通过了！${NC}"
|
||||
exit 0
|
||||
else
|
||||
echo -e "${RED}部分测试失败${NC}"
|
||||
exit 1
|
||||
fi
|
||||
}
|
||||
|
||||
# 显示帮助信息
|
||||
show_help() {
|
||||
echo "使用方法: $0 [模型类型] [测试步骤]"
|
||||
echo
|
||||
echo "参数:"
|
||||
echo " 模型类型: onnx, caffe, tensorflow, paddle, all (默认: onnx)"
|
||||
echo " 测试步骤: 01_load, 02_quantize, 03_quantize_hybrid, 04_add_pre_post,"
|
||||
echo " 05_export, 06_inference, 07_inference_hybrid, 08_dump,"
|
||||
echo " 09_add_pre_post, all (默认: all)"
|
||||
echo
|
||||
echo "示例:"
|
||||
echo " $0 onnx all # 测试所有ONNX模型，运行完整流程"
|
||||
echo " $0 all 01_load # 所有模型类型，只运行load步骤"
|
||||
echo " $0 tensorflow 05_export # 测试TensorFlow模型，只运行export步骤"
|
||||
}
|
||||
|
||||
# 处理特殊参数
|
||||
if [ "$1" = "-h" ] || [ "$1" = "--help" ]; then
|
||||
show_help
|
||||
exit 0
|
||||
fi
|
||||
|
||||
# 运行主函数
|
||||
main
|
||||
|
|
@ -0,0 +1,54 @@
|
|||
#!/bin/bash
|
||||
# 简化版测试脚本：演示如何处理examples目录下的模型
|
||||
# 使用方法: ./test_examples_simple.sh
|
||||
|
||||
set -e
|
||||
|
||||
# 颜色输出
|
||||
RED='\033[0;31m'
|
||||
GREEN='\033[0;32m'
|
||||
YELLOW='\033[1;33m'
|
||||
NC='\033[0m'
|
||||
|
||||
echo -e "${YELLOW}=== Netrans Examples 简化测试开始 ===${NC}"
|
||||
|
||||
# 基础路径
|
||||
EXAMPLES_DIR="/home/xj/work/nudt/netrans/examples"
|
||||
TEST_DIR="/home/xj/work/nudt/netrans/test/netrans_cli"
|
||||
|
||||
# 查找ONNX模型
|
||||
echo -e "${YELLOW}查找ONNX模型...${NC}"
|
||||
models=$(find "$EXAMPLES_DIR" -name "*.onnx" -type f | head -3)
|
||||
|
||||
if [ -z "$models" ]; then
|
||||
echo -e "${RED}未找到ONNX模型${NC}"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
echo -e "${GREEN}找到以下模型:${NC}"
|
||||
echo "$models" | nl
|
||||
echo
|
||||
|
||||
# 测试第一个模型
|
||||
first_model=$(echo "$models" | head -1)
|
||||
model_dir=$(dirname "$first_model")
|
||||
model_name=$(basename "$first_model" .onnx)
|
||||
|
||||
echo -e "${YELLOW}测试第一个模型: $model_name${NC}"
|
||||
echo -e "模型目录: $model_dir"
|
||||
echo
|
||||
|
||||
# 运行load测试
|
||||
echo -e "${YELLOW}运行 01_load.sh 测试...${NC}"
|
||||
cd "$TEST_DIR"
|
||||
if bash 01_load.sh "$model_dir"; then
|
||||
echo -e "${GREEN}✓ Load 测试通过${NC}"
|
||||
else
|
||||
echo -e "${RED}✗ Load 测试失败${NC}"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
echo
|
||||
echo -e "${GREEN}=== 简化测试完成 ===${NC}"
|
||||
echo -e "提示: 可以编辑此脚本测试其他模型或步骤"
|
||||
echo -e "完整版脚本: $TEST_DIR/run_examples_test.sh"
|
||||
|
|
@ -0,0 +1,37 @@
|
|||
# 第三方依赖
|
||||
|
||||
此目录存放项目依赖的外部 whl 包。
|
||||
|
||||
## 文件列表
|
||||
|
||||
| 文件 | 版本 | 说明 |
|
||||
|------|------|------|
|
||||
| acuity-6.33.19-cp310-cp310-manylinux2010_x86_64.whl | 6.33.19 | PNNA 芯片模型编译核心库 |
|
||||
|
||||
## 获取方式
|
||||
|
||||
acuity whl 文件请联系项目维护团队获取。
|
||||
|
||||
## 使用说明
|
||||
|
||||
### 开发环境安装
|
||||
|
||||
```bash
|
||||
# 使用 --no-deps 避免 acuity 覆盖已安装的 torch 版本
|
||||
# torch 已在 requirements.txt 中单独安装
|
||||
pip install vendor/acuity-6.33.19-cp310-cp310-manylinux2010_x86_64.whl --no-deps
|
||||
```
|
||||
|
||||
### 离线打包
|
||||
|
||||
构建离线包时会自动包含此目录的 whl 文件：
|
||||
|
||||
```bash
|
||||
cd devtools/packing/
|
||||
python3 download_dependencies.py # 会自动复制 vendor/ 中的 acuity
|
||||
bash build.sh
|
||||
```
|
||||
|
||||
## 版本历史
|
||||
|
||||
- **6.33.19** (2026-02-09) - 当前版本
|
||||
Loading…
Reference in New Issue