<div align="center">
<img src="./img/logo.png"/ width=600>
</div>
<!--
<div align="center">
✨ <em>"The path to AGI is continual, INFTY paves the way."</em> ✨ </strong><br />
<em>— INFTY × AGI</em>
</div>
-->
<!--
[![Typing SVG](https://readme-typing-svg.demolab.com?font=Fira+Code&size=30&duration=7000&pause=1000&random=false&width=1500&height=100&lines=✨"The+path+to+AGI+is+continual,INFTY+paves+the+way."+✨+—+INFTY+×+AGI)](https://git.io/typing-svg)
-->
<p align="center">
<img src="https://readme-typing-svg.demolab.com?font=Fira+Code&size=30¢er=true&duration=7000&pause=1000&random=false&width=1500&height=80&lines=%E2%9C%A8%22The+path+to+AGI+is+continual,INFTY+paves+the+way.%22%E2%9C%A8+%E2%80%94+INFTY+%C3%97+AGI" />
</p>
-----------
<p align="center">
<a href="https://github.com/THUDM/INFTY/"><img src="https://img.shields.io/badge/Documentation-Read%20the%20Docs-blue.svg" alt="Documentation"></a>
<a href="https://arxiv.org/abs/2501.01045"><img src="https://img.shields.io/badge/arXiv-2507.12345-b31b1b.svg" alt="arXiv"></a>
<a href="https://pytorch.org/get-started/locally/"><img alt="PyTorch" src="https://img.shields.io/badge/PyTorch-ee4c2c?logo=pytorch&logoColor=white"></a>
<a href="https://pypi.org/project/numpy/1.24.4/"><img src="https://img.shields.io/badge/numpy-1.24.4-blue.svg" alt="numpy"></a>
<a href="https://pypi.org/project/matplotlib/3.7.5/"><img src="https://img.shields.io/badge/matplotlib-3.7.5-blue.svg" alt="matplotlib"></a>
<a href="https://opensource.org/licenses/MIT"><img src="https://img.shields.io/badge/License-MIT-green.svg" alt="License: MIT"></a>
</p>
<div align="center">
<center><h2>INFTY Engine: An Optimization Toolkit to Support Continual AI</h2></center>
</div>
# 🌈 What is INFTY?
<div align="center">
<img src="./img/INFTY_demo.gif" alt="animated" />
</div>
</br>
**INFTY**, a flexible and user-friendly optimization engine tailored for Continual AI. INFTY includes a suite of built-in optimization algorithms that directly tackle core challenges (e.g., catastrophic forgetting, stability–plasticity dilemma, generalization) in Continual AI. INFTY supports plug-and-play and theoretical analysis utilities, compatible with: i) various Continual AI, e.g., PTM-based CL, and Continual PEFT, Continual Diffusion, and Continual VLM etc.; ii) diverse models, e.g., ResNet, Transformer, ViT, CLIP, and Diffusion. INFTY provides a unified optimization solution in Continual AI, can serve as infrastructure for broad deployment.
# ✨ Features
- **Generality**: Built-in CL–friendly optimization algorithms, supporting a wide range of scenarios, models, methods, and learning paradigms.
- **Usability**: Portable, plugin-style design, enabling easy replacement of fixed options within existing pipelines.
- **Utilities**: Built-in tools for theoretical analysis and visualization, facilitating investigation and diagnostic insight into optimization behavior.
# 🧠 Algorithms
INFTY has implemented three mainstream algorithms currently:
<div align="center">
<img src="./img/overview.png"/>
</div>
# 📚 Versatile Case (Ongoing Updates)
## Scenario 1: Typical Continual Learning
### Case 1: Generalizability support
This category covers standard sharpness-aware generalization methods for exploring flatter regions. These methods can be applied to most architectures and training platforms, either from scratch or with pre-trained models (PTMs). Details can be found in [C_Flat](https://arxiv.org/abs/2404.00986v1).
<div align="center">
<img src="./img/case1.png"/>
</div>
### Case 2: BP-Free support
This category focuses on gradient approximation when backpropagation is not feasible. Combining with PTMs is strongly recommended to achieve better initialization and faster convergence. Details can be found in [ZeroFlow](https://arxiv.org/abs/2501.01045).
<div align="center">
<img src="./img/case2.png"/>
</div>
### Case 3: Multi-objective support
This category mitigates gradient interference between old and new task objectives, with gradient manipulation applied solely to shared parameters. Details can be found in [UniGrad_FS](https://ieeexplore.ieee.org/abstract/document/10636267).
<div align="center">
<img src="./img/case3.png"/>
</div>
## Scenario 2: Continual Text-to-Image Diffusion Model
INFTY empowers CIDM! A tiny demo shows how INFTY can be applied to train Concept-Incremental text-to-image Diffusion Models. Origin repo can be found in [CIDM](https://github.com/JiahuaDong/CIFC).
<div align="center">
<img src="./img/cidm.png"/>
</div>
## Scenario 3: Vision-Language Continual Learning
INFTY also supports multi-modal continual learning — ready for VLMs, AVLMs, and more. Origin repo can be found in [DMNSP](https://github.com/RL-VIG/DMNSP).
| Method | T1 | T2 | T3 | T4 | T5 | T6 | T7 | T8 | T9 | T10 | Avg |
| -------- | :-----: | :-----: | :-----: | :-----: | :-----: | :-----: | :-----: | :-----: | :-----: | :-----: | :-----: |
| DMNSP | **99.20** | 96.10 | **91.93** | 87.05 | 87.00 | 86.10 | 84.17 | 83.05 | 81.58 | 79.94 | 87.61 |
|+INFTY | **99.20** | **96.30** | 91.80 | **87.30** | **87.44** | **86.60** | **84.46** | **83.20** | **81.69** | **80.52** | **87.85** |
# 🛠️ Installation
## Option 1: Using pip
```bash
pip install infty
```
## Option 2: Install from source
```bash
conda create -n infty python=3.8
conda activate infty
git clone https://github.com/THUDM/INFTY.git
cd infty && pip install .
```
# 🚀 Quick start
Thanks to the PILOT repo, we provide a simple example showcasing INFTY Engine. Hyperparameters for specific methods are configured in `../infty_configs/`.
```bash
cd infty
pip install .[examples]
cd examples/PILOT
python main.py --config=exps/memo_scr.json --inftyopt=c_flat
python main.py --config=exps/ease.json --inftyopt=zo_sgd_conserve
python main.py --config=exps/icarl.json --inftyopt=unigrad_fs
```
Tips: Feel free to use INFTY in your own projects following 🛠️ Installation or 🧩 Custom usage.
# 🧩 Custom usage
## Optimizers
Step 1. Wrap your base optimizer with an INFTY optimizer
```python
from infty import optim as infty_optim
base_optimizer = optim.SGD(
filter(lambda p: p.requires_grad, self._network.parameters()),
lr=self.args['lrate'],
momentum=0.9,
weight_decay=self.args['weight_decay']
)
optimizer = infty_optim.C_Flat(params=self._network.parameters(), base_optimizer=base_optimizer, model=self._network, args=self.args)
```
Step 2. Implement the create_loss_fn function
```python
def create_loss_fn(self, inputs, targets):
"""
Create a closure to calculate the loss
"""
def loss_fn():
outputs = self._network(inputs)
logits = outputs["logits"]
loss_clf = F.cross_entropy(logits, targets)
return logits, [loss_clf]
return loss_fn
```
Step 3. Use the loss_fn to calculate the loss and backward
```python
loss_fn = self.create_loss_fn(inputs, targets)
optimizer.set_closure(loss_fn)
logits, loss_list = optimizer.step()
```
## Visualization plots
INFTY includes built-in visualization tools for inspecting optimization behavior:
- [x] **Loss Landscape**: visualize sharpness around local minima
- [x] **Hessian ESD**: curvature analysis via eigenvalue spectrum density
- [x] **Conflict Curves**: quantify gradient interference (supports PCGrad, GradVac, UniGrad_FS, CAGrad)
- [x] **Optimization Trajectory**: observe optimization directions under gradient shifts with a toy example
```python
from infty import plot as infty_plot
infty_plot.visualize_landscape(self._network, self.create_loss_fn, train_loader, self._cur_task, self._device)
infty_plot.visualize_esd(self._network, self.create_loss_fn, train_loader, self._cur_task, self._device)
infty_plot.visualize_conflicts()
infty_plot.visualize_trajectory(optim="c_flat")
```
# 📝 Citation
If any content in this repo is useful for your work, please cite the following paper:
- `ZeroFlow:` Zeroflow: Overcoming catastrophic forgetting is easier than you think. ICML 2025 [[paper]](https://arxiv.org/abs/2501.01045)
- `C-Flat++:` C-Flat++: Towards a More Efficient and Powerful Framework for Continual Learning. Arxiv 2025 [[paper]](https://arxiv.org/abs/2508.18860)
- `C-Flat:` Make Continual Learning Stronger via C-Flat. NeurIPS 2024 [[paper]](https://arxiv.org/abs/2404.00986v1)
- `UniGrad-FS:` UniGrad-FS: Unified Gradient Projection With Flatter Sharpness for Continual Learning. TII 2024 [[paper]](https://ieeexplore.ieee.org/abstract/document/10636267)
# 🙏 Acknowledgements
We thank the following repos providing helpful components/functions in our work.
- [PILOT](https://github.com/LAMDA-CL/LAMDA-PILOT) /
[PyCIL](https://github.com/LAMDA-CL/PyCIL) /
[CIDM](https://github.com/JiahuaDong/CIFC) /
[DMNSP](https://github.com/RL-VIG/DMNSP) /
[GAM](https://github.com/xxgege/GAM) /
[ZO-LLM](https://github.com/ZO-Bench/ZO-LLM) /
[CAGrad](https://github.com/Cranial-XIX/CAGrad) /
[PyHessian](https://github.com/amirgholami/PyHessian)
# 📬 Contact us
If you have any questions, feel free to open an issue or contact the corresponding authors: Wei Li (ymjiii98@gmail.com) or Tao Feng (fengtao.hi@gmail.com).
# 🧾 License
This project is licensed under the MIT License.
Raw data
{
"_id": null,
"home_page": "https://github.com/THUDM/INFTY",
"name": "infty",
"maintainer": null,
"docs_url": null,
"requires_python": ">=3.6",
"maintainer_email": null,
"keywords": "optimization, visualization, machine learning, deep learning",
"author": "taofeng weili",
"author_email": "taofeng weili <ymjiii98@gmail.com>",
"download_url": "https://files.pythonhosted.org/packages/24/34/29784f1cbd1c24aacc01f436441b97d61ee203fc5f814f48a6787b00ec8c/infty-0.1.1.tar.gz",
"platform": null,
"description": "<div align=\"center\">\n<img src=\"./img/logo.png\"/ width=600> \n</div>\n\n<!--\n<div align=\"center\">\n\u2728 <em>\"The path to AGI is continual, INFTY paves the way.\"</em> \u2728 </strong><br />\n<em>\u2014 INFTY \u00d7 AGI</em>\n</div>\n-->\n\n<!--\n[![Typing SVG](https://readme-typing-svg.demolab.com?font=Fira+Code&size=30&duration=7000&pause=1000&random=false&width=1500&height=100&lines=\u2728\"The+path+to+AGI+is+continual,INFTY+paves+the+way.\"+\u2728+\u2014+INFTY+\u00d7+AGI)](https://git.io/typing-svg)\n-->\n\n<p align=\"center\">\n <img src=\"https://readme-typing-svg.demolab.com?font=Fira+Code&size=30¢er=true&duration=7000&pause=1000&random=false&width=1500&height=80&lines=%E2%9C%A8%22The+path+to+AGI+is+continual,INFTY+paves+the+way.%22%E2%9C%A8+%E2%80%94+INFTY+%C3%97+AGI\" />\n</p>\n\n-----------\n\n<p align=\"center\">\n <a href=\"https://github.com/THUDM/INFTY/\"><img src=\"https://img.shields.io/badge/Documentation-Read%20the%20Docs-blue.svg\" alt=\"Documentation\"></a>\n <a href=\"https://arxiv.org/abs/2501.01045\"><img src=\"https://img.shields.io/badge/arXiv-2507.12345-b31b1b.svg\" alt=\"arXiv\"></a>\n <a href=\"https://pytorch.org/get-started/locally/\"><img alt=\"PyTorch\" src=\"https://img.shields.io/badge/PyTorch-ee4c2c?logo=pytorch&logoColor=white\"></a>\n <a href=\"https://pypi.org/project/numpy/1.24.4/\"><img src=\"https://img.shields.io/badge/numpy-1.24.4-blue.svg\" alt=\"numpy\"></a>\n <a href=\"https://pypi.org/project/matplotlib/3.7.5/\"><img src=\"https://img.shields.io/badge/matplotlib-3.7.5-blue.svg\" alt=\"matplotlib\"></a>\n <a href=\"https://opensource.org/licenses/MIT\"><img src=\"https://img.shields.io/badge/License-MIT-green.svg\" alt=\"License: MIT\"></a>\n</p>\n\n<div align=\"center\">\n <center><h2>INFTY Engine: An Optimization Toolkit to Support Continual AI</h2></center>\n</div>\n\n# \ud83c\udf08 What is INFTY?\n\n<div align=\"center\">\n<img src=\"./img/INFTY_demo.gif\" alt=\"animated\" />\n</div>\n</br>\n\n**INFTY**, a flexible and user-friendly optimization engine tailored for Continual AI. INFTY includes a suite of built-in optimization algorithms that directly tackle core challenges (e.g., catastrophic forgetting, stability\u2013plasticity dilemma, generalization) in Continual AI. INFTY supports plug-and-play and theoretical analysis utilities, compatible with: i) various Continual AI, e.g., PTM-based CL, and Continual PEFT, Continual Diffusion, and Continual VLM etc.; ii) diverse models, e.g., ResNet, Transformer, ViT, CLIP, and Diffusion. INFTY provides a unified optimization solution in Continual AI, can serve as infrastructure for broad deployment.\n\n\n# \u2728 Features\n- **Generality**: Built-in CL\u2013friendly optimization algorithms, supporting a wide range of scenarios, models, methods, and learning paradigms.\n \n- **Usability**: Portable, plugin-style design, enabling easy replacement of fixed options within existing pipelines.\n \n- **Utilities**: Built-in tools for theoretical analysis and visualization, facilitating investigation and diagnostic insight into optimization behavior.\n\n# \ud83e\udde0 Algorithms\nINFTY has implemented three mainstream algorithms currently:\n\n<div align=\"center\">\n<img src=\"./img/overview.png\"/> \n</div>\n\n\n# \ud83d\udcda Versatile Case (Ongoing Updates)\n## Scenario 1: Typical Continual Learning\n### Case 1: Generalizability support \nThis category covers standard sharpness-aware generalization methods for exploring flatter regions. These methods can be applied to most architectures and training platforms, either from scratch or with pre-trained models (PTMs). Details can be found in [C_Flat](https://arxiv.org/abs/2404.00986v1).\n<div align=\"center\">\n<img src=\"./img/case1.png\"/> \n</div>\n\n### Case 2: BP-Free support\nThis category focuses on gradient approximation when backpropagation is not feasible. Combining with PTMs is strongly recommended to achieve better initialization and faster convergence. Details can be found in [ZeroFlow](https://arxiv.org/abs/2501.01045).\n<div align=\"center\">\n<img src=\"./img/case2.png\"/> \n</div>\n\n### Case 3: Multi-objective support\nThis category mitigates gradient interference between old and new task objectives, with gradient manipulation applied solely to shared parameters. Details can be found in [UniGrad_FS](https://ieeexplore.ieee.org/abstract/document/10636267).\n<div align=\"center\">\n<img src=\"./img/case3.png\"/> \n</div>\n\n## Scenario 2: Continual Text-to-Image Diffusion Model\nINFTY empowers CIDM! A tiny demo shows how INFTY can be applied to train Concept-Incremental text-to-image Diffusion Models. Origin repo can be found in [CIDM](https://github.com/JiahuaDong/CIFC).\n<div align=\"center\">\n<img src=\"./img/cidm.png\"/> \n</div>\n\n\n\n## Scenario 3: Vision-Language Continual Learning\nINFTY also supports multi-modal continual learning \u2014 ready for VLMs, AVLMs, and more. Origin repo can be found in [DMNSP](https://github.com/RL-VIG/DMNSP).\n\n| Method | T1 | T2 | T3 | T4 | T5 | T6 | T7 | T8 | T9 | T10 | Avg |\n| -------- | :-----: | :-----: | :-----: | :-----: | :-----: | :-----: | :-----: | :-----: | :-----: | :-----: | :-----: |\n| DMNSP | **99.20** | 96.10 | **91.93** | 87.05 | 87.00 | 86.10 | 84.17 | 83.05 | 81.58 | 79.94 | 87.61 |\n|+INFTY | **99.20** | **96.30** | 91.80 | **87.30** | **87.44** | **86.60** | **84.46** | **83.20** | **81.69** | **80.52** | **87.85** | \n\n\n\n# \ud83d\udee0\ufe0f Installation\n\n## Option 1: Using pip\n\n```bash\npip install infty\n```\n## Option 2: Install from source\n```bash\nconda create -n infty python=3.8\n\nconda activate infty\n\ngit clone https://github.com/THUDM/INFTY.git\n\ncd infty && pip install .\n```\n\n# \ud83d\ude80 Quick start\nThanks to the PILOT repo, we provide a simple example showcasing INFTY Engine. Hyperparameters for specific methods are configured in `../infty_configs/`. \n```bash\ncd infty \n\npip install .[examples]\n\ncd examples/PILOT\n\npython main.py --config=exps/memo_scr.json --inftyopt=c_flat\npython main.py --config=exps/ease.json --inftyopt=zo_sgd_conserve\npython main.py --config=exps/icarl.json --inftyopt=unigrad_fs\n```\nTips: Feel free to use INFTY in your own projects following \ud83d\udee0\ufe0f Installation or \ud83e\udde9 Custom usage.\n\n# \ud83e\udde9 Custom usage\n## Optimizers\nStep 1. Wrap your base optimizer with an INFTY optimizer\n```python\nfrom infty import optim as infty_optim\n\nbase_optimizer = optim.SGD(\n filter(lambda p: p.requires_grad, self._network.parameters()), \n lr=self.args['lrate'], \n momentum=0.9, \n weight_decay=self.args['weight_decay']\n )\noptimizer = infty_optim.C_Flat(params=self._network.parameters(), base_optimizer=base_optimizer, model=self._network, args=self.args)\n```\nStep 2. Implement the create_loss_fn function\n```python\ndef create_loss_fn(self, inputs, targets):\n \"\"\"\n Create a closure to calculate the loss\n \"\"\"\n def loss_fn():\n outputs = self._network(inputs)\n logits = outputs[\"logits\"]\n loss_clf = F.cross_entropy(logits, targets)\n return logits, [loss_clf]\n return loss_fn\n```\nStep 3. Use the loss_fn to calculate the loss and backward\n```python\nloss_fn = self.create_loss_fn(inputs, targets)\noptimizer.set_closure(loss_fn)\nlogits, loss_list = optimizer.step()\n```\n\n## Visualization plots\nINFTY includes built-in visualization tools for inspecting optimization behavior:\n- [x] **Loss Landscape**: visualize sharpness around local minima\n- [x] **Hessian ESD**: curvature analysis via eigenvalue spectrum density\n- [x] **Conflict Curves**: quantify gradient interference (supports PCGrad, GradVac, UniGrad_FS, CAGrad)\n- [x] **Optimization Trajectory**: observe optimization directions under gradient shifts with a toy example\n```python\nfrom infty import plot as infty_plot\n\ninfty_plot.visualize_landscape(self._network, self.create_loss_fn, train_loader, self._cur_task, self._device)\ninfty_plot.visualize_esd(self._network, self.create_loss_fn, train_loader, self._cur_task, self._device)\ninfty_plot.visualize_conflicts()\ninfty_plot.visualize_trajectory(optim=\"c_flat\")\n```\n\n# \ud83d\udcdd Citation\nIf any content in this repo is useful for your work, please cite the following paper:\n\n- `ZeroFlow:` Zeroflow: Overcoming catastrophic forgetting is easier than you think. ICML 2025 [[paper]](https://arxiv.org/abs/2501.01045)\n\n- `C-Flat++:` C-Flat++: Towards a More Efficient and Powerful Framework for Continual Learning. Arxiv 2025 [[paper]](https://arxiv.org/abs/2508.18860)\n\n- `C-Flat:` Make Continual Learning Stronger via C-Flat. NeurIPS 2024 [[paper]](https://arxiv.org/abs/2404.00986v1)\n\n- `UniGrad-FS:` UniGrad-FS: Unified Gradient Projection With Flatter Sharpness for Continual Learning. TII 2024 [[paper]](https://ieeexplore.ieee.org/abstract/document/10636267)\n\n# \ud83d\ude4f Acknowledgements\nWe thank the following repos providing helpful components/functions in our work.\n\n- [PILOT](https://github.com/LAMDA-CL/LAMDA-PILOT) /\n[PyCIL](https://github.com/LAMDA-CL/PyCIL) /\n[CIDM](https://github.com/JiahuaDong/CIFC) /\n[DMNSP](https://github.com/RL-VIG/DMNSP) /\n[GAM](https://github.com/xxgege/GAM) /\n[ZO-LLM](https://github.com/ZO-Bench/ZO-LLM) /\n[CAGrad](https://github.com/Cranial-XIX/CAGrad) /\n[PyHessian](https://github.com/amirgholami/PyHessian)\n\n# \ud83d\udcec Contact us\nIf you have any questions, feel free to open an issue or contact the corresponding authors: Wei Li (ymjiii98@gmail.com) or Tao Feng (fengtao.hi@gmail.com).\n\n# \ud83e\uddfe License\nThis project is licensed under the MIT License.\n",
"bugtrack_url": null,
"license": "MIT",
"summary": "A continual learning optimizer and visualization toolkit",
"version": "0.1.1",
"project_urls": {
"Homepage": "https://github.com/THUDM/INFTY"
},
"split_keywords": [
"optimization",
" visualization",
" machine learning",
" deep learning"
],
"urls": [
{
"comment_text": null,
"digests": {
"blake2b_256": "bea1dfbccf000f2936fc66b9e59875c901036b9fcc723af418991a97bad70c65",
"md5": "0d839eec348e88076e70c5d4c6542af6",
"sha256": "62d39c5237bb3c27c1c633629df7a8f8a72911ef626efdb52c76387fd1e12160"
},
"downloads": -1,
"filename": "infty-0.1.1-py3-none-any.whl",
"has_sig": false,
"md5_digest": "0d839eec348e88076e70c5d4c6542af6",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.6",
"size": 42770,
"upload_time": "2025-09-11T08:51:27",
"upload_time_iso_8601": "2025-09-11T08:51:27.079389Z",
"url": "https://files.pythonhosted.org/packages/be/a1/dfbccf000f2936fc66b9e59875c901036b9fcc723af418991a97bad70c65/infty-0.1.1-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": null,
"digests": {
"blake2b_256": "243429784f1cbd1c24aacc01f436441b97d61ee203fc5f814f48a6787b00ec8c",
"md5": "2eefa003f2e2705827ad11acf51b720d",
"sha256": "75e8f3ba4fe81f6e9a74a56670366da57952cedbea26939b5a70e4972b97c3ec"
},
"downloads": -1,
"filename": "infty-0.1.1.tar.gz",
"has_sig": false,
"md5_digest": "2eefa003f2e2705827ad11acf51b720d",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.6",
"size": 26958623,
"upload_time": "2025-09-11T08:51:44",
"upload_time_iso_8601": "2025-09-11T08:51:44.814479Z",
"url": "https://files.pythonhosted.org/packages/24/34/29784f1cbd1c24aacc01f436441b97d61ee203fc5f814f48a6787b00ec8c/infty-0.1.1.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2025-09-11 08:51:44",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "THUDM",
"github_project": "INFTY",
"github_not_found": true,
"lcname": "infty"
}
JSON
