torchvista - GitLite

KO | EN

gitlite — search

#hacktoberfest #ai #typescript #llm #android #rust #docker #macos #nodejs #awesome #cli #awesome-list #security #open-source

torchvista

(★ 759)

Interactive Pytorch forward pass visualization in notebooks

파일 탐색기

최종 버전 다운로드 (.zip)

- - collapsible-graph.gif
  - demo.gif
  - error-graph.png
  - info-popup.png
  - interactive-graph.gif
- - tutorial.html
- demos.html
- index.html
- tutorials.html
- examples_generator.py
- find_function_mode_overrides.py
- __init__.py
- _helpers.py
- test_build_immediate_ancestor_map.py
- test_extract_tensors_from_obj.py
- test_process_graph.py
- test_transform_to_nested_graph.py
- test_validate_export_format.py
- - __init__.py
  - d3.min.js
  - jsoneditor-10.2.0.min.css
  - jsoneditor-10.2.0.min.js
  - viz-standalone.js
- - __init__.py
  - demo-graph.html
  - graph.html
- __init__.py
- engine.py
- enums.py
- graph_transforms.py
- module_discovery.py
- overrides.py
- render.py
- tensor_utils.py
- tracer.py
.gitignore
LICENSE
NOTICE.txt
README.md
requirements.txt
setup.py

# 설치 가이드

1. 코드 내려받기

git clone https://github.com/sachinhosmani/torchvista

깃허브에서 프로젝트 코드 전체를 내 컴퓨터로 내려받습니다.

cd torchvista

방금 내려받은 프로젝트 폴더 안으로 이동합니다.

2. 공식 설치 스크립트

쉬움 추천

사전 준비물

Python 3 pip 명령어를 쓰려면 Python이 필요합니다.

pip install torchvista

PyPI에 배포된 패키지를 바로 설치합니다. 소스 클론이 필요 없습니다.

✅ 설치 후 새 터미널을 열고, 프로그램의 버전 확인 명령(예: --version)으로 정상 설치됐는지 확인하세요.

이 레포의 README에 적힌 실제 명령어를 그대로 가져왔습니다.

3. Python

쉬움

사전 준비물

Git GitHub에서 프로젝트 코드를 내려받으려면 필요합니다.
Python 3 설치 시 'Add Python to PATH' 옵션을 꼭 체크하세요 (Windows).

pip install torchvista

PyPI에 배포된 패키지를 바로 설치합니다. 소스 클론이 필요 없습니다.

pip install -e ".[test]"

requirements.txt 등에 명시된 파이썬 라이브러리를 설치합니다.

✅ 에러 메시지 없이 실행되고 터미널에 안내 문구가 출력되면 정상입니다.

이 레포의 README에 적힌 실제 명령어를 그대로 가져왔습니다.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# torchvista

An interactive tool to visualize the forward pass of a PyTorch model directly in the notebook—with a single line of code. Works with web-based notebooks like Jupyter, Google Colab and Kaggle. Also allows you to export the visualization as image, svg and HTML.

## ✨ Features

### Interactive graph with drag and zoom support

![](https://raw.githubusercontent.com/sachinhosmani/torchvista/main/docs/assets/interactive-graph.gif)

--------

### Collapsible nodes for hierarchical modules

![](https://raw.githubusercontent.com/sachinhosmani/torchvista/main/docs/assets/collapsible-graph.gif)

--------

### Error-tolerant partial visualization when errors arise
(e.g., shape mismatches) for ease of debugging

![](https://raw.githubusercontent.com/sachinhosmani/torchvista/main/docs/assets/error-graph.png)

--------

### Click on nodes to view parameter and attribute info

![](https://raw.githubusercontent.com/sachinhosmani/torchvista/main/docs/assets/info-popup.png)

--------

## Tutorials and examples

- Step-by-step tutorial 👉 [here](https://sachinhosmani.github.io/torchvista/tutorials.html)
- Quick Google Colab tutorial 👉 [here](https://colab.research.google.com/drive/1wrWKhpvGiqHhE0Lb1HnFGeOcS4uBqGXw?usp=sharing) (must be logged in to Colab)
- Check out demos 👉 [here](https://sachinhosmani.github.io/torchvista/demos.html)

## How does it work?

I wrote a technical deep-dive blog post on Towards Data Science which can be read [here](https://towardsdatascience.com/torchvista-building-an-interactive-pytorch-visualization-package-for-notebooks/). It explains how the tool works under the hood.

## ⚙️ Usage

Install via pip
```
pip install torchvista
```

Alternatively, install via `conda` (See [`torchvista-feedstock`](https://github.com/conda-forge/torchvista-feedstock?tab=readme-ov-file#installing-torchvista) for more information)

```
conda install -c conda-forge torchvista
```

Run from your **web-based notebook** (Jupyter, Colab, VSCode notebook, etc)

```
import torch
import torch.nn as nn

# Import torchvista
from torchvista import trace_model

# Define your module
class LinearModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(10, 5)

def forward(self, x):
        return self.linear(x)

# Instantiate the module and tensor input
model = LinearModel()
inputs = torch.randn(2, 10)
model.eval() # Optional

# Trace!
trace_model(model, inputs)
```
## API: `trace_model`

```python
trace_model(
    model,
    inputs,
    show_non_gradient_nodes=True,
    collapse_modules_after_depth=1,
    forced_module_tracing_depth=None,
    height=800,
    width=None,
    export_format=None,
    show_module_attr_names=False,
    export_path=None,
    show_compressed_view=False,
)
```

| Parameter | Type | Default Value | Category | Description |
| --- | --- | --- | --- | --- |
| `model` | `torch.nn.Module` | — | Tracing | Model instance to visualize. |
| `inputs` | `Any` | — | Tracing | Input(s) forwarded into the model; pass a single input or a tuple. |
| `show_non_gradient_nodes` | `bool` | `True` | Visual | Display nodes for constants and other values outside the gradient graph. |
| `collapse_modules_after_depth` | `int` | `1` | Visual | Depth to initially expand nested modules; `0` collapses everything (nodes can still be expanded interactively). |
| `forced_module_tracing_depth` | `int` | `None` | Tracing | Maximum depth of module internals to trace; `None` traces only user-defined modules. |
| `height` | `int` | `800` | Visual | Canvas height in pixels. |
| `width` | `int \| str` | `None` | Visual | Canvas width; accepts pixels or percentages; defaults to full available width when omitted. |
| `export_format` | `str` | `None` | Export | Optional export format: `png`, `svg`, or `html` if exporting graph as a file. Otherwise, by default the graph is shown within the notebook.|
| `show_module_attr_names` | `bool` | `False` | Visual | Display attribute names for modules when available instead of just class names. |
| `export_path` | `str` | `None` | Export | Custom path if exporting as a file. **Only HTML format** is currently supported with custom export paths. If only file name is specified, it will be created inside the present working directory. |
| `show_compressed_view` (Experimental) | `bool` | `False` | Visual | Compress the graph by showing repeating nodes of the same type with identical input and output dims in single "repeat" blocks. This feature currently only recognises repeating nodes within `Sequential` and `ModuleList`. WARNING: this feature might be expensive on large models. |

## Running tests

Tests live under [`tests/`](tests/) and use [pytest](https://docs.pytest.org/).

```bash
pip install -e ".[test]"
pytest
```