nv-generate-vae-finetune

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

NV-Generate-VAE-Finetune

Purpose

用途

Used for finetuning the NV-Generate-CTMR MAISI VAE/autoencoder from user-supplied CT or MRI NIfTI training volumes.
Not for clinical interpretation, regulatory use, or approving synthetic data for production training.
Upstream currently documents VAE training in
```
train_vae_tutorial.ipynb
```
and provides configs/helpers, but not a
```
scripts.train_vae
```
CLI. This skill does not execute the notebook; it stages the required config/datalist glue locally and uses upstream helper APIs.

Manifest I/O: inputs are

datalist

and

data_base_dir

; outputs are

autoencoder_checkpoint

discriminator_checkpoint

, and

result_json

The underlying training contract is the upstream config/env JSON (
```
config_maisi_vae_train.json
```
+
```
environment_maisi_vae_train.json
```
, as used in
```
train_vae_tutorial.ipynb
```
). The wrapper stages those JSON files for you and exposes the most-tuned fields as CLI flags; the sections below document the fields, their defaults, and how to monitor/tune a run.

用于基于用户提供的CT或MRI NIfTI训练数据集微调NV-Generate-CTMR MAISI VAE/自动编码器。
不用于临床解读、合规性使用，也不用于批准合成数据用于生产训练。
上游项目目前在
```
train_vae_tutorial.ipynb
```
中记录了VAE训练流程，并提供配置文件/辅助工具，但未提供
```
scripts.train_vae
```
命令行界面（CLI）。本技能不执行该notebook；它会在本地准备所需的配置/数据列表关联文件，并使用上游的辅助API。

清单输入输出：输入为

datalist

和

data_base_dir

；输出为

autoencoder_checkpoint

、

discriminator_checkpoint

和

result_json

。

底层训练规则遵循上游的配置/环境JSON文件（
```
config_maisi_vae_train.json
```
+
```
environment_maisi_vae_train.json
```
，与
```
train_vae_tutorial.ipynb
```
中使用的一致）。该封装工具会为您准备这些JSON文件，并将最常用的可配置字段暴露为CLI参数；以下章节将记录这些字段、默认值以及如何监控/调优训练过程。

Instructions

操作说明

Read
```
skill_manifest.yaml
```
before changing arguments, side effects, or validation gates.
Run
```
scripts/run_vae_finetune.py
```
from the Medical AI Skills repo root.
If a host agent exposes
```
run_script
```
, use
```
run_script("scripts/run_vae_finetune.py", args=[...])
```
; otherwise run the Bash/Python command below.
Use
```
--preflight
```
first when checking a new datalist; remove
```
--preflight
```
only when the user explicitly wants to launch GPU finetuning.
For a staged preflight input bundle directory, use
```
BUNDLE/preflight_datalist.json
```
as the datalist and
```
BUNDLE/preflight_dataset
```
as
```
--data-base-dir
```
when those files are present.

在修改参数、副作用或验证规则前，请先阅读
```
skill_manifest.yaml
```
。
从Medical AI Skills仓库根目录运行
```
scripts/run_vae_finetune.py
```
。
如果宿主Agent提供
```
run_script
```
方法，请使用
```
run_script("scripts/run_vae_finetune.py", args=[...])
```
；否则运行下方的Bash/Python命令。
检查新数据列表时，请先使用
```
--preflight
```
参数；仅当用户明确希望启动GPU微调时，再移除该参数。
对于已准备好的预检查输入包目录，当存在相关文件时，使用
```
BUNDLE/preflight_datalist.json
```
作为数据列表，
```
BUNDLE/preflight_dataset
```
作为
```
--data-base-dir
```
。

Examples

示例

Validate and stage a preflight finetune check from an input bundle (the recommended first step — no GPU, no training). This is the single canonical command; replace

INPUT_BUNDLE

and

OUT_DIR

with your paths:

bash

export NV_GENERATE_ROOT="${NV_GENERATE_ROOT:-.workbench_data/upstreams/NV-Generate-CTMR}" && \
python skills/nv-generate-vae-finetune/scripts/run_vae_finetune.py \
  INPUT_BUNDLE/preflight_datalist.json \
  --data-base-dir INPUT_BUNDLE/preflight_dataset \
  --output-dir OUT_DIR \
  --modality mri \
  --preflight

For real GPU finetuning and other variations, see Usage below.

从输入包中验证并准备预检查微调（推荐的第一步——无需GPU，不进行训练）。这是标准命令；请将

INPUT_BUNDLE

和

OUT_DIR

替换为您的路径：

bash

export NV_GENERATE_ROOT="${NV_GENERATE_ROOT:-.workbench_data/upstreams/NV-Generate-CTMR}" && \
python skills/nv-generate-vae-finetune/scripts/run_vae_finetune.py \
  INPUT_BUNDLE/preflight_datalist.json \
  --data-base-dir INPUT_BUNDLE/preflight_dataset \
  --output-dir OUT_DIR \
  --modality mri \
  --preflight

如需进行真实的GPU微调或其他操作变体，请参阅下方的【2. 使用方式（一键训练）】章节。

Available Scripts

可用脚本

Script	Purpose	Arguments
`scripts/run_vae_finetune.py`	Primary entrypoint declared by `skill_manifest.yaml` .	`DATALIST.json --data-base-dir DATA_DIR --output-dir OUT_DIR [--epochs N] [--modality mri] [--patch-size 64,64,64] [--preflight]`

脚本	用途	参数
`scripts/run_vae_finetune.py`	`skill_manifest.yaml` 中声明的主入口脚本	`DATALIST.json --data-base-dir DATA_DIR --output-dir OUT_DIR [--epochs N] [--modality mri] [--patch-size 64,64,64] [--preflight]`

Prerequisites

前置条件

NV_GENERATE_ROOT

may point to a current checkout of

https://github.com/NVIDIA-Medtech/NV-Generate-CTMR

containing

configs/config_maisi_vae_train.json

scripts/transforms.py

, and

scripts/utils.py

NV_GENERATE_ROOT

is unset, the wrapper searches

.workbench_data/upstreams/NV-Generate-CTMR

```
CUDA_VISIBLE_DEVICES
```
is optional and can be used to select the GPU for real training.
Runtime requirements: NVIDIA CUDA GPU for real training, Python packages from the upstream
```
requirements.txt
```
,
```
lpips
```
, and downloaded VAE weights unless using
```
--train-from-scratch
```
.
Side effects: writes staged configs, checkpoints, TensorBoard logs, and run summaries under the caller-provided
```
--output-dir
```
; may write model caches under the upstream checkout,
```
~/.cache/huggingface/
```
, and
```
~/.cache/torch/
```
; may contact
```
https://huggingface.co
```
,
```
https://github.com
```
, and
```
https://download.pytorch.org
```
.
The datalist is a MONAI-style JSON object with non-empty
```
training[]
```
and
```
validation[]
```
or
```
testing[]
```
. Each entry has an
```
image
```
path relative to
```
--data-base-dir
```
and optional
```
class
```
or
```
modality
```
of
```
ct
```
or
```
mri
```
.

NV_GENERATE_ROOT

需指向

https://github.com/NVIDIA-Medtech/NV-Generate-CTMR

的当前检出版本，其中需包含

configs/config_maisi_vae_train.json

、

scripts/transforms.py

和

scripts/utils.py

。

如果未设置

NV_GENERATE_ROOT

，封装工具会搜索

.workbench_data/upstreams/NV-Generate-CTMR

。

```
CUDA_VISIBLE_DEVICES
```
为可选参数，可用于选择真实训练时使用的GPU。
运行时要求：真实训练需NVIDIA CUDA GPU；需安装上游
```
requirements.txt
```
中的Python包、
```
lpips
```
，以及下载的VAE权重（除非使用
```
--train-from-scratch
```
参数）。
副作用：会在用户指定的
```
--output-dir
```
下写入准备好的配置文件、检查点、TensorBoard日志和运行摘要；可能会在上游检出目录、
```
~/.cache/huggingface/
```
和
```
~/.cache/torch/
```
下写入模型缓存；可能会连接
```
https://huggingface.co
```
、
```
https://github.com
```
和
```
https://download.pytorch.org
```
。
数据列表为MONAI风格的JSON对象，需包含非空的
```
training[]
```
和
```
validation[]
```
或
```
testing[]
```
。每个条目需包含相对于
```
--data-base-dir
```
的
```
image
```
路径，以及可选的
```
class
```
或
```
modality
```
（值为
```
ct
```
或
```
mri
```
）。

1. Config and environment JSON (adapt to your data)

1. 配置与环境JSON（适配您的数据）

The wrapper copies the upstream VAE config/env JSON from

$NV_GENERATE_ROOT/configs

, rewrites the fields below, and writes the staged copies under

OUT_DIR/workflow/configs/

. You normally only set your datalist and data root; the listed CLI flags override individual fields when you need to.

Environment JSON (

environment_maisi_vae_train.json

Field	Set from	Notes
`model_dir`	`--output-dir`	Where `autoencoder.pt` / `discriminator.pt` and best checkpoints are saved.
`tfevent_path`	`--output-dir`	TensorBoard event directory.
`finetune`	`--train-from-scratch`	`true` (default) loads `trained_autoencoder_path` ; the flag sets it `false` .
`trained_autoencoder_path`	upstream weights / `--trained-autoencoder-path`	Starting VAE checkpoint when finetuning.

Training fields (

config_maisi_vae_train.json

Field	Flag	Type	Default	Notes
`autoencoder_train.n_epochs`	`--epochs`	int	`1`
`autoencoder_train.batch_size`	`--batch-size`	int	`1`	Per-GPU (single-GPU runner).
`autoencoder_train.patch_size`	`--patch-size`	int,int,int	`64,64,64`	Training crop.
`autoencoder_train.val_batch_size`	`--val-batch-size`	int	`1`
`autoencoder_train.val_sliding_window_patch_size`	`--val-sliding-window-patch-size`	int,int,int	`96,96,64`	Sliding-window validation ROI.
`autoencoder_train.lr`	`--lr`	float	`1e-4`
`autoencoder_train.perceptual_weight`	`--perceptual-weight`	float	`0.3`	LPIPS term.
`autoencoder_train.kl_weight`	`--kl-weight`	float	`1e-7`	KL term.
`autoencoder_train.adv_weight`	`--adv-weight`	float	`0.1`	Adversarial term.
`autoencoder_train.recon_loss`	`--recon-loss`	`l1` \| `l2`	`l1`
`autoencoder_train.val_interval`	`--val-interval`	int	`1`	Epochs between validation passes.
`autoencoder_train.cache`	`--cache-rate`	float	`0.0`	MONAI `CacheDataset` fraction.
`autoencoder_train.amp`	`--no-amp`	flag	on	Mixed precision; flag disables it.
`data_option.random_aug`	`--no-random-aug`	flag	on	Random augmentation; flag disables it.
`data_option.spacing_type`	`--spacing-type`	`original` \| `fixed` \| `rand_zoom`	`original`
`data_option.spacing`	`--spacing`	float,float,float	unset	Required when `spacing_type` is `fixed` / `rand_zoom` .
`data_option.select_channel`	`--select-channel`	int	`0`	Channel for multi-channel inputs.

--modality

(

ct

mri

, default

mri

) fills the per-entry

class

for datalist items missing one. Validation/testing entries are required because the training loop runs a validation pass.

For an end-to-end reference including example data download, see the upstream tutorial

train_vae_tutorial.ipynb

封装工具会从

$NV_GENERATE_ROOT/configs

复制上游的VAE配置/环境JSON，重写以下字段，并将准备好的副本写入

OUT_DIR/workflow/configs/

。通常您只需设置数据列表和数据根目录；当需要时，可通过列出的CLI参数覆盖单个字段。

环境JSON（

environment_maisi_vae_train.json

）：

字段	来源	说明
`model_dir`	`--output-dir`	保存 `autoencoder.pt` / `discriminator.pt` 和最优检查点的目录。
`tfevent_path`	`--output-dir`	TensorBoard事件目录。
`finetune`	`--train-from-scratch`	默认值为 `true` （加载 `trained_autoencoder_path` ）；该参数会将其设置为 `false` 。
`trained_autoencoder_path`	上游权重 / `--trained-autoencoder-path`	微调时使用的初始VAE检查点。

训练字段（

config_maisi_vae_train.json

）：

字段	参数	类型	默认值	说明
`autoencoder_train.n_epochs`	`--epochs`	int	`1`
`autoencoder_train.batch_size`	`--batch-size`	int	`1`	单GPU运行时的每GPU批次大小。
`autoencoder_train.patch_size`	`--patch-size`	int,int,int	`64,64,64`	训练时的裁剪尺寸。
`autoencoder_train.val_batch_size`	`--val-batch-size`	int	`1`
`autoencoder_train.val_sliding_window_patch_size`	`--val-sliding-window-patch-size`	int,int,int	`96,96,64`	滑动窗口验证的感兴趣区域（ROI）。
`autoencoder_train.lr`	`--lr`	float	`1e-4`
`autoencoder_train.perceptual_weight`	`--perceptual-weight`	float	`0.3`	LPIPS项权重。
`autoencoder_train.kl_weight`	`--kl-weight`	float	`1e-7`	KL项权重。
`autoencoder_train.adv_weight`	`--adv-weight`	float	`0.1`	对抗项权重。
`autoencoder_train.recon_loss`	`--recon-loss`	`l1` \| `l2`	`l1`
`autoencoder_train.val_interval`	`--val-interval`	int	`1`	两次验证之间的训练轮数。
`autoencoder_train.cache`	`--cache-rate`	float	`0.0`	MONAI `CacheDataset` 的缓存比例。
`autoencoder_train.amp`	`--no-amp`	参数	开启	混合精度训练；该参数会关闭此功能。
`data_option.random_aug`	`--no-random-aug`	参数	开启	随机数据增强；该参数会关闭此功能。
`data_option.spacing_type`	`--spacing-type`	`original` \| `fixed` \| `rand_zoom`	`original`
`data_option.spacing`	`--spacing`	float,float,float	未设置	当 `spacing_type` 为 `fixed` / `rand_zoom` 时为必填项。
`data_option.select_channel`	`--select-channel`	int	`0`	多通道输入时选择的通道。

--modality

（值为

ct

或

mri

，默认

mri

）会为数据列表中缺少

class

的条目填充该字段。必须提供验证/测试条目，因为训练循环会执行验证步骤。

如需包含示例数据下载的端到端参考，请参阅上游教程

train_vae_tutorial.ipynb

。

2. Usage (one-line training)

2. 使用方式（一键训练）

Preflight only:

bash

export NV_GENERATE_ROOT="${NV_GENERATE_ROOT:-.workbench_data/upstreams/NV-Generate-CTMR}" && \
python skills/nv-generate-vae-finetune/scripts/run_vae_finetune.py \
  PATH_TO_DATALIST.json \
  --data-base-dir PATH_TO_DATA_ROOT \
  --output-dir runs/nv_generate_vae_finetune_preflight \
  --preflight

Preflight bundle input:

bash

export NV_GENERATE_ROOT="${NV_GENERATE_ROOT:-.workbench_data/upstreams/NV-Generate-CTMR}" && \
python skills/nv-generate-vae-finetune/scripts/run_vae_finetune.py \
  PATH_TO_INPUT_BUNDLE/preflight_datalist.json \
  --data-base-dir PATH_TO_INPUT_BUNDLE/preflight_dataset \
  --output-dir runs/nv_generate_vae_finetune_preflight \
  --preflight

GPU finetuning:

bash

export NV_GENERATE_ROOT="${NV_GENERATE_ROOT:-.workbench_data/upstreams/NV-Generate-CTMR}" && \
python -m pip install -r "$NV_GENERATE_ROOT/requirements.txt" && \
python -m pip install lpips tensorboard && \
python skills/nv-generate-vae-finetune/scripts/run_vae_finetune.py \
  PATH_TO_DATALIST.json \
  --data-base-dir PATH_TO_DATA_ROOT \
  --output-dir runs/nv_generate_vae_finetune \
  --epochs 1 \
  --modality mri \
  --patch-size 64,64,64 \
  --download-model-data

Replace

PATH_TO_DATALIST.json

and

PATH_TO_DATA_ROOT

with the user's actual paths. Do not use the fixture datalist for real training; it is a preflight-only placeholder.

仅预检查：

bash

export NV_GENERATE_ROOT="${NV_GENERATE_ROOT:-.workbench_data/upstreams/NV-Generate-CTMR}" && \
python skills/nv-generate-vae-finetune/scripts/run_vae_finetune.py \
  PATH_TO_DATALIST.json \
  --data-base-dir PATH_TO_DATA_ROOT \
  --output-dir runs/nv_generate_vae_finetune_preflight \
  --preflight

预检查包输入：

bash

export NV_GENERATE_ROOT="${NV_GENERATE_ROOT:-.workbench_data/upstreams/NV-Generate-CTMR}" && \
python skills/nv-generate-vae-finetune/scripts/run_vae_finetune.py \
  PATH_TO_INPUT_BUNDLE/preflight_datalist.json \
  --data-base-dir PATH_TO_INPUT_BUNDLE/preflight_dataset \
  --output-dir runs/nv_generate_vae_finetune_preflight \
  --preflight

GPU微调：

bash

export NV_GENERATE_ROOT="${NV_GENERATE_ROOT:-.workbench_data/upstreams/NV-Generate-CTMR}" && \
python -m pip install -r "$NV_GENERATE_ROOT/requirements.txt" && \
python -m pip install lpips tensorboard && \
python skills/nv-generate-vae-finetune/scripts/run_vae_finetune.py \
  PATH_TO_DATALIST.json \
  --data-base-dir PATH_TO_DATA_ROOT \
  --output-dir runs/nv_generate_vae_finetune \
  --epochs 1 \
  --modality mri \
  --patch-size 64,64,64 \
  --download-model-data

请将

PATH_TO_DATALIST.json

和

PATH_TO_DATA_ROOT

替换为用户的实际路径。请勿使用测试数据列表进行真实训练；它仅作为预检查的占位符。

3. Monitor training (TensorBoard)

3. 监控训练（TensorBoard）

The runner writes TensorBoard scalars (per-iteration and per-epoch

recons_loss

kl_loss

p_loss

, adversarial/real/fake losses, and a validation

scale_factor

) under

OUT_DIR/artifacts/tfevent/autoencoder

. Launch TensorBoard against the output directory:

bash

python -m pip install tensorboard && \
tensorboard --logdir runs/nv_generate_vae_finetune/artifacts/tfevent

The same per-epoch loss history is also captured in

OUT_DIR/artifacts/workflow_summary.json

and echoed in the JSON the wrapper prints to stdout (

loss_history

, best-checkpoint paths,

exit_code

stderr_tail

运行器会在

OUT_DIR/artifacts/tfevent/autoencoder

下写入TensorBoard标量数据（每迭代和每轮的

recons_loss

、

kl_loss

、

p_loss

、对抗/真实/虚假损失，以及验证

scale_factor

）。针对输出目录启动TensorBoard：

bash

python -m pip install tensorboard && \
tensorboard --logdir runs/nv_generate_vae_finetune/artifacts/tfevent

相同的每轮损失历史也会被记录在

OUT_DIR/artifacts/workflow_summary.json

中，并在封装工具打印到标准输出的JSON中回显（包含

loss_history

、最优检查点路径、

exit_code

、

stderr_tail

）。

4. Hyperparameter tuning and common pitfalls

4. 超参数调优与常见问题

Reconstructions blurry — raise
```
--perceptual-weight
```
(default
```
0.3
```
); try
```
--recon-loss l2
```
if edges look washed out.
Posterior collapse / over-regularized latents —
```
--kl-weight
```
is intentionally tiny (
```
1e-7
```
); increasing it too much degrades reconstruction.
Adversarial training unstable — lower
```
--adv-weight
```
(default
```
0.1
```
) or
```
--lr
```
; a warmup schedule already ramps the LR over the first 20 epochs.

Out-of-memory — reduce

--patch-size

(e.g.

48,48,48

) and

--val-sliding-window-patch-size

, keep

--batch-size 1

, and lower

--cache-rate

datalist must include non-empty validation[] or testing[]
— the validation loop is mandatory; add

validation[]

(or

testing[]

) entries.

Single-GPU only — the runner asserts exactly one CUDA GPU; set
```
CUDA_VISIBLE_DEVICES
```
to pick which one.

重建结果模糊 —— 提高
```
--perceptual-weight
```
（默认值
```
0.3
```
）；如果边缘看起来模糊，尝试使用
```
--recon-loss l2
```
。
后验坍缩 / 潜在空间过度正则化 ——
```
--kl-weight
```
默认设置得非常小（
```
1e-7
```
）；过度增大该值会降低重建质量。
对抗训练不稳定 —— 降低
```
--adv-weight
```
（默认值
```
0.1
```
）或
```
--lr
```
；学习率预热调度已在前20轮逐步提升学习率。

内存不足 —— 减小

--patch-size

（例如

48,48,48

）和

--val-sliding-window-patch-size

，保持

--batch-size 1

，并降低

--cache-rate

。

datalist must include non-empty validation[] or testing[]
—— 验证循环是必须的；添加

validation[]

（或

testing[]

）条目。

仅支持单GPU —— 运行器会检查是否存在且仅存在一个CUDA GPU；设置
```
CUDA_VISIBLE_DEVICES
```
来选择使用哪一个。

5. Evaluate the finetuned VAE

5. 评估微调后的VAE

Validation reconstruction loss (lowest-

val_weighted_loss

epoch) is tracked automatically and the best autoencoder is saved as

autoencoder_epochN.pt

under

OUT_DIR/artifacts/models

. To evaluate downstream:

Compare validation
```
recons_loss
```
/
```
p_loss
```
curves across runs in TensorBoard, and
Plug the finetuned autoencoder into a diffusion finetune/generation run (e.g.
```
nv-generate-mr-brain-finetune
```
via
```
--trained-autoencoder-path
```
) to confirm latents still decode to usable volumes.

This skill gates file accounting and reconstruction bookkeeping only — image quality and downstream utility must be judged by a domain expert.

验证重建损失（

val_weighted_loss

最低的轮次）会被自动跟踪，最优自动编码器会保存为

OUT_DIR/artifacts/models

下的

autoencoder_epochN.pt

。如需进行下游评估：

在TensorBoard中比较不同运行的验证
```
recons_loss
```
/
```
p_loss
```
曲线，并且
将微调后的自动编码器插入到扩散模型微调/生成运行中（例如通过
```
--trained-autoencoder-path
```
参数传入
```
nv-generate-mr-brain-finetune
```
），以确认潜在空间仍能解码为可用的数据集。

本技能仅负责文件管理和重建记录——图像质量和下游效用必须由领域专家判断。

Limitations

局限性

Requires a current upstream
```
NV-Generate-CTMR
```
checkout with VAE configs and helper APIs. The skill owns the runner glue and does not depend on the notebook.
Full training can be expensive and is not deterministic across hardware, CUDA, and package versions.
The wrapper gates file accounting and command provenance, not anatomical realism, reconstruction quality, or downstream model utility.
Not for clinical deployment, clinical interpretation, autonomous diagnosis, regulatory submission, or production training-data approval.

需要当前版本的上游
```
NV-Generate-CTMR
```
检出版本，包含VAE配置文件和辅助API。本技能负责运行器的封装逻辑，不依赖于notebook。
完整训练可能成本高昂，且在不同硬件、CUDA和包版本下结果不具有确定性。
封装工具仅负责文件管理和命令溯源，不保证解剖学真实性、重建质量或下游模型效用。
不用于临床部署、临床解读、自主诊断、合规提交或生产训练数据审批。

Troubleshooting

故障排除

Error	Cause	Fix
`VAE configs/helpers were not found`	`NV_GENERATE_ROOT` does not point at a current NV-Generate-CTMR checkout.	Clone or update `https://github.com/NVIDIA-Medtech/NV-Generate-CTMR` and set `NV_GENERATE_ROOT` .
`datalist must include non-empty validation[] or testing[]`	VAE training requires validation data for the configured validation loop.	Add `validation[]` or `testing[]` entries with relative image paths.
CUDA, MONAI, or LPIPS import failure	Runtime environment lacks upstream dependencies.	Install `"$NV_GENERATE_ROOT/requirements.txt"` plus `lpips tensorboard` in the selected environment.

错误	原因	解决方法
`VAE configs/helpers were not found`	`NV_GENERATE_ROOT` 未指向当前版本的NV-Generate-CTMR检出目录。	克隆或更新 `https://github.com/NVIDIA-Medtech/NV-Generate-CTMR` 并设置 `NV_GENERATE_ROOT` 。
`datalist must include non-empty validation[] or testing[]`	VAE训练需要验证数据以执行配置好的验证循环。	添加包含相对图像路径的 `validation[]` 或 `testing[]` 条目。
CUDA、MONAI或LPIPS导入失败	运行环境缺少上游依赖。	在所选环境中安装 `"$NV_GENERATE_ROOT/requirements.txt"` 中的包以及 `lpips tensorboard` 。