黑芝麻A1000 docker环境中使用 bsnntool 转换模型（六）

1. 查看《BST-AI工具链快速上手指南-v0.4.pdf》，下载并解压镜像文件，进入 /xxx/bsnn_tools-v3.7.0/bsnn_tools ,修改 load.sh 脚本，主要修改这三处：：

PORT=8370
# 本地PC端目录
MODEL_ZOO_DIR=/xxx/heizhima/nn/model_zoo_dir
OPTION_NOTEBOOK=" nohup jupyter notebook --ip=0.0.0.0 --port=8888 --allow-root --NotebookApp.contents_manager_class='notedown.NotedownContentsManager' &"

2. 如果之前进入过docker环境中，退出后又进入，这时会出现问题：

Error response from daemon: Container 9ee317b36456a779c1c1055b34c2de22ff7f59d517083bd0b15fc0ce6ad46875 is not running

这时这样做

$ sudo docker ps -a
[sudo] password for wxf: 
CONTAINER ID        IMAGE                 COMMAND                  CREATED             STATUS                  PORTS                    NAMES
9ee317b36456        bsnn_tools:3.7.0      "nohup jupyter noteb…"   4 days ago          Exited (0) 3 days ago                            bsnn_tools_root
ced9f8c97c3b        portainer/portainer   "/portainer"             2 weeks ago         Up About an hour        0.0.0.0:9000->9000/tcp   portainer

$ sudo docker stop 9ee317b36456
9ee317b36456
$ sudo docker rm 9ee317b36456

然后再运行

$ sudo ./load.sh
$ sudo docker ps -a
CONTAINER ID        IMAGE                 COMMAND                  CREATED             STATUS              PORTS                    NAMES
1f18c9a461ca        bsnn_tools:3.7.0      "nohup jupyter noteb…"   45 seconds ago      Up 41 seconds       0.0.0.0:8370->8888/tcp   bsnn_tools_root
ced9f8c97c3b        portainer/portainer   "/portainer"             2 weeks ago         Up About an hour    0.0.0.0:9000->9000/tcp   portainer
$ sudo docker exec -it 1f18c9a461ca /bin/bash

3. 打开浏览器输入http://0.0.0.0:8370，登录 jupyter

root@1f18c9a461ca:/workspace# jupyter notebook list
Currently running servers:
http://0.0.0.0:8888/?token=7798cc7ca9c38c309863fa25eeffb5b97da3299c1b2cfcaf :: /workspace

复制 **token=**后面的字符 7798cc7ca9c38c309863fa25eeffb5b97da3299c1b2cfcaf 各自的不一样，输入 jupyter 页面中的 **Password or token:**后面，登录进入

进入之后，如图

4. 宿主机中，复制模型到共享目录
   脚本 load.sh 中的 MODEL_ZOO_DIR=/xxx/heizhima/nn/model_zoo_dir 目录是docker环境和宿主机的共享目录，把 /xxx/nn/bsnn_tools-v3.7.0/bsnn_tools/model 目录下的 data 、test1 文件夹复制到 MODEL_ZOO_DIR 目录下，这时 docker 环境中 /workspace/models 目录下有

root@1f18c9a461ca:/workspace/models# ls
data  test1

5. 进入docker环境，使用 bsnntool 转换模型
   查看 /workspace/models/test1/data/test1.yaml 中的路径是否正确

# cat test1/data/test1.yaml
model_path: /workspace/models/test1/data/mini_model.tf.pb
input_data_set_path: /workspace/models/data/imagenet_mini
data_reader_method: image_folder_data_reader
image_process_config: /workspace/models/data/imagenet_mini_process_config.json
model_name: test1_mini_model
model_description:
  link: www.samplelink.com
  description: This is a test mini model
  author: bst
  date: '2021-03-08'
stage:
  - stage_name: pre_processing_stage
    priority: 100
  - stage_name: graph_optimization_stage
    priority: 200
  - stage_name: quantization_stage
    priority: 300
  - stage_name: graph_partition_stage
    priority: 400
  - stage_name: section_binding_stage
    priority: 500
  - stage_name: code_generation_stage
    priority: 600
  - stage_name: code_compilation_stage
    priority: 700
  - stage_name: run_emulation_stage
    profiling_mode: 2
    priority: 800
  - stage_name: run_emulation_stage
    profiling_mode: 0
    priority: 1000
  - stage_name: report_generation_stage
    priority: 1200

root@1f18c9a461ca:/workspace# bstnnx_run --config /workspace/models/test1/data/test1.yaml --result /workspace/bstnnx_test_result/test1
root@1f18c9a461ca:/workspace# cd bstnnx_test_result/test1/
root@1f18c9a461ca:/workspace/bstnnx_test_result/test1# ls
1000_RunEmulationStage      200_GraphOptimizationStage  500_SectionBindingStage   800_RunEmulationStage
100_PreProcessingStage      300_QuantizationStage       600_CodeGenerationStage   job.log
1200_ReportGenerationStage  400_GraphPartitionStage     700_CodeCompilationStage  main_flow.yaml

jupyter 中显示