Metadata-Version: 2.4
Name: dolphinflow
Version: 0.2.0
Summary: DolphinFlow: A robust, low-maintenance optimizer for PyTorch.
Author-email: Eric Hartford <ehartford@gmail.com>
License:                                  Apache License
                                   Version 2.0, January 2004
                                http://www.apache.org/licenses/
        
           TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
        
           1. Definitions.
        
              "License" shall mean the terms and conditions for use, reproduction,
              and distribution as defined by Sections 1 through 9 of this document.
        
              "Licensor" shall mean the copyright owner or entity authorized by
              the copyright owner that is granting the License.
        
              "Legal Entity" shall mean the union of the acting entity and all
              other entities that control, are controlled by, or are under common
              control with that entity. For the purposes of this definition,
              "control" means (i) the power, direct or indirect, to cause the
              direction or management of such entity, whether by contract or
              otherwise, or (ii) ownership of fifty percent (50%) or more of the
              outstanding shares, or (iii) beneficial ownership of such entity.
        
              "You" (or "Your") shall mean an individual or Legal Entity
              exercising permissions granted by this License.
        
              "Source" form shall mean the preferred form for making modifications,
              including but not limited to software source code, documentation
              source, and configuration files.
        
              "Object" form shall mean any form resulting from mechanical
              transformation or translation of a Source form, including but
              not limited to compiled object code, generated documentation,
              and conversions to other media types.
        
              "Work" shall mean the work of authorship, whether in Source or
              Object form, made available under the License, as indicated by a
              copyright notice that is included in or attached to the work
              (an example is provided in the Appendix below).
        
              "Derivative Works" shall mean any work, whether in Source or Object
              form, that is based on (or derived from) the Work and for which the
              editorial revisions, annotations, elaborations, or other modifications
              represent, as a whole, an original work of authorship. For the purposes
              of this License, Derivative Works shall not include works that remain
              separable from, or merely link (or bind by name) to the interfaces of,
              the Work and Derivative Works thereof.
        
              "Contribution" shall mean any work of authorship, including
              the original version of the Work and any modifications or additions
              to that Work or Derivative Works thereof, that is intentionally
              submitted to Licensor for inclusion in the Work by the copyright owner
              or by an individual or Legal Entity authorized to submit on behalf of
              the copyright owner. For the purposes of this definition, "submitted"
              means any form of electronic, verbal, or written communication sent
              to the Licensor or its representatives, including but not limited to
              communication on electronic mailing lists, source code control systems,
              and issue tracking systems that are managed by, or on behalf of, the
              Licensor for the purpose of discussing and improving the Work, but
              excluding communication that is conspicuously marked or otherwise
              designated in writing by the copyright owner as "Not a Contribution."
        
              "Contributor" shall mean Licensor and any individual or Legal Entity
              on behalf of whom a Contribution has been received by Licensor and
              subsequently incorporated within the Work.
        
           2. Grant of Copyright License. Subject to the terms and conditions of
              this License, each Contributor hereby grants to You a perpetual,
              worldwide, non-exclusive, no-charge, royalty-free, irrevocable
              copyright license to reproduce, prepare Derivative Works of,
              publicly display, publicly perform, sublicense, and distribute the
              Work and such Derivative Works in Source or Object form.
        
           3. Grant of Patent License. Subject to the terms and conditions of
              this License, each Contributor hereby grants to You a perpetual,
              worldwide, non-exclusive, no-charge, royalty-free, irrevocable
              (except as stated in this section) patent license to make, have made,
              use, offer to sell, sell, import, and otherwise transfer the Work,
              where such license applies only to those patent claims licensable
              by such Contributor that are necessarily infringed by their
              Contribution(s) alone or by combination of their Contribution(s)
              with the Work to which such Contribution(s) was submitted. If You
              institute patent litigation against any entity (including a
              cross-claim or counterclaim in a lawsuit) alleging that the Work
              or a Contribution incorporated within the Work constitutes direct
              or contributory patent infringement, then any patent licenses
              granted to You under this License for that Work shall terminate
              as of the date such litigation is filed.
        
           4. Redistribution. You may reproduce and distribute copies of the
              Work or Derivative Works thereof in any medium, with or without
              modifications, and in Source or Object form, provided that You
              meet the following conditions:
        
              (a) You must give any other recipients of the Work or
                  Derivative Works a copy of this License; and
        
              (b) You must cause any modified files to carry prominent notices
                  stating that You changed the files; and
        
              (c) You must retain, in the Source form of any Derivative Works
                  that You distribute, all copyright, patent, trademark, and
                  attribution notices from the Source form of the Work,
                  excluding those notices that do not pertain to any part of
                  the Derivative Works; and
        
              (d) If the Work includes a "NOTICE" text file as part of its
                  distribution, then any Derivative Works that You distribute must
                  include a readable copy of the attribution notices contained
                  within such NOTICE file, excluding those notices that do not
                  pertain to any part of the Derivative Works, in at least one
                  of the following places: within a NOTICE text file distributed
                  as part of the Derivative Works; within the Source form or
                  documentation, if provided along with the Derivative Works; or,
                  within a display generated by the Derivative Works, if and
                  wherever such third-party notices normally appear. The contents
                  of the NOTICE file are for informational purposes only and
                  do not modify the License. You may add Your own attribution
                  notices within Derivative Works that You distribute, alongside
                  or as an addendum to the NOTICE text from the Work, provided
                  that such additional attribution notices cannot be construed
                  as modifying the License.
        
              You may add Your own copyright statement to Your modifications and
              may provide additional or different license terms and conditions
              for use, reproduction, or distribution of Your modifications, or
              for any such Derivative Works as a whole, provided Your use,
              reproduction, and distribution of the Work otherwise complies with
              the conditions stated in this License.
        
           5. Submission of Contributions. Unless You explicitly state otherwise,
              any Contribution intentionally submitted for inclusion in the Work
              by You to the Licensor shall be under the terms and conditions of
              this License, without any additional terms or conditions.
              Notwithstanding the above, nothing herein shall supersede or modify
              the terms of any separate license agreement you may have executed
              with Licensor regarding such Contributions.
        
           6. Trademarks. This License does not grant permission to use the trade
              names, trademarks, service marks, or product names of the Licensor,
              except as required for reasonable and customary use in describing the
              origin of the Work and reproducing the content of the NOTICE file.
        
           7. Disclaimer of Warranty. Unless required by applicable law or
              agreed to in writing, Licensor provides the Work (and each
              Contributor provides its Contributions) on an "AS IS" BASIS,
              WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
              implied, including, without limitation, any warranties or conditions
              of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
              PARTICULAR PURPOSE. You are solely responsible for determining the
              appropriateness of using or redistributing the Work and assume any
              risks associated with Your exercise of permissions under this License.
        
           8. Limitation of Liability. In no event and under no legal theory,
              whether in tort (including negligence), contract, or otherwise,
              unless required by applicable law (such as deliberate and grossly
              negligent acts) or agreed to in writing, shall any Contributor be
              liable to You for damages, including any direct, indirect, special,
              incidental, or consequential damages of any character arising as a
              result of this License or out of the use or inability to use the
              Work (including but not limited to damages for loss of goodwill,
              work stoppage, computer failure or malfunction, or any and all
              other commercial damages or losses), even if such Contributor
              has been advised of the possibility of such damages.
        
           9. Accepting Warranty or Additional Liability. While redistributing
              the Work or Derivative Works thereof, You may choose to offer,
              and charge a fee for, acceptance of support, warranty, indemnity,
              or other liability obligations and/or rights consistent with this
              License. However, in accepting such obligations, You may act only
              on Your own behalf and on Your sole responsibility, not on behalf
              of any other Contributor, and only if You agree to indemnify,
              defend, and hold each Contributor harmless for any liability
              incurred by, or claims asserted against, such Contributor by reason
              of your accepting any such warranty or additional liability.
        
           END OF TERMS AND CONDITIONS
        
           APPENDIX: How to apply the Apache License to your work.
        
              To apply the Apache License to your work, attach the following
              boilerplate notice, with the fields enclosed by brackets "[]"
              replaced with your own identifying information. (Don't include
              the brackets!)  The text should be enclosed in the appropriate
              comment syntax for the file format. We also recommend that a
              file or class name and description of purpose be included on the
              same "printed page" as the copyright notice for easier
              identification within third-party archives.
        
           Copyright [yyyy] [name of copyright owner]
        
           Licensed under the Apache License, Version 2.0 (the "License");
           you may not use this file except in compliance with the License.
           You may obtain a copy of the License at
        
               http://www.apache.org/licenses/LICENSE-2.0
        
           Unless required by applicable law or agreed to in writing, software
           distributed under the License is distributed on an "AS IS" BASIS,
           WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
           See the License for the specific language governing permissions and
           limitations under the License.
Project-URL: Homepage, https://github.com/cognitivecomputations/dolphinflow-optimizer
Project-URL: Bug Tracker, https://github.com/cognitivecomputations/dolphinflow-optimizer/issues
Keywords: pytorch,optimizer,machine learning,deep learning
Classifier: Development Status :: 4 - Beta
Classifier: Intended Audience :: Developers
Classifier: Intended Audience :: Science/Research
Classifier: License :: OSI Approved :: MIT License
Classifier: Operating System :: OS Independent
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.8
Classifier: Programming Language :: Python :: 3.9
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Requires-Python: >=3.8
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: torch>=1.10
Provides-Extra: bnb
Requires-Dist: bitsandbytes>=0.41.0; extra == "bnb"
Dynamic: license-file

# DolphinFlow Optimizer

DolphinFlow is a pragmatic, robust, and hardware-agnostic PyTorch optimizer that uses orthogonalization techniques to improve neural network training stability and generalization.

DolphinFlow is about results. I'm a practitioner, not a theoretician. I build things that do things. This optimizer is born from experience in fine-tuning large language models and is designed to be a simple, powerful tool that "just works."

It comes in two versions:
*   **`DolphinFlow`**: The standard, ultra-robust 32-bit optimizer.
*   **`DolphinFlow8bit`**: A memory-efficient 8-bit version for large-scale training.

## Core Idea: Why Orthogonalization?

Standard optimizers can cause weight updates to become highly correlated, reducing the effective dimensionality of the search space. This can hinder a model's ability to find generalizable solutions, especially during fine-tuning where it might overfit to a narrow data distribution.

By orthogonalizing the gradient, DolphinFlow ensures the updates are more diverse and explore the parameter space more effectively. The `ortho_mode="vector"` setting is particularly effective at preventing **Naïve Loss Minimization**—a phenomenon where the model simply scales up its weights without actually learning new features. This idea is heavily inspired by the analysis of grokking and numerical stability in modern deep learning.

## Installation

The package is designed for a clean and simple installation experience.

### Standard (32-bit) Version

The standard version has no dependencies beyond PyTorch.

```bash
pip install dolphinflow-optimizer
```

### 8-bit Version (Recommended for Large Models)

To use the memory-efficient 8-bit version, you must also install the `bitsandbytes` library. You can do this by specifying the `[bnb]` extra during installation:

```bash
pip install dolphinflow-optimizer[bnb]
```

## Basic Usage

### Using the Standard `DolphinFlow`

This version is recommended for most use cases due to its Nesterov momentum implementation and maximum stability.

```python
import torch
from dolphinflow import DolphinFlow

# Example model
model = torch.nn.Linear(100, 10)

# Use it like any other PyTorch optimizer
optimizer = DolphinFlow(model.parameters(), lr=1e-4)
```

### Using the 8-bit `DolphinFlow8bit`

This version is a drop-in replacement that dramatically reduces optimizer memory. **Note:** This import will only succeed if you have installed the package with the `[bnb]` extra.

```python
import torch
from dolphinflow import DolphinFlow8bit

model = torch.nn.Linear(100, 10)

# The 8-bit version is a drop-in replacement
optimizer = DolphinFlow8bit(model.parameters(), lr=1e-4)
```

## Key Features & Parameters

The API has been simplified to its essential, robust components.

*   `lr: float = 1e-4`: The learning rate. A low LR is essential for fine-tuning.
*   `ortho_mode: str = "vector"`: The orthogonalization strategy.
    *   **`"vector"`** (Default): Recommended for all use cases. Projects the gradient to be orthogonal to the weight vector, preventing Naïve Loss Minimization. Works for all layer types.
    *   **`"matrix"`**: Applies a Newton-Schulz iteration to the entire weight matrix. More computationally intensive and best for smaller 2D layers.
    *   **`None`**: Disables orthogonalization entirely.
*   `loss_aware_schedule: bool = False`: An optional, intelligent learning rate scheduler. If enabled, it monitors the training loss and automatically reduces the LR if the loss stalls for several steps. **Requires passing a `closure` to `optimizer.step()`**.
*   `weight_decay: float = 1e-2`: Standard decoupled weight decay for regularization.
*   `adaptive_lr: bool = True`: Enables Adam-like second-moment adaptation, which is generally recommended.
*   `gradient_clipping: float = 1.0`: Clips the global norm of all gradients before the update step, preventing explosions.

## Advanced Usage: Combining with StableMax

The `grokking` paper by Prieto et al., which inspired the `ortho_mode="vector"` feature, identified two related issues in training dynamics:
1.  **The Cause:** Naïve Loss Minimization (NLM), where the optimizer scales up weights, leading to uncontrolled logit growth.
2.  **The Symptom:** Softmax Collapse, a numerical instability that occurs when logits become too large for standard floating-point precision.

`DolphinFlow` and `StableMax` form a powerful synergistic pair to address both issues:
*   **`DolphinFlow`** with `ortho_mode="vector"` addresses the **cause** by preventing the NLM gradient direction.
*   **`StableMax`** addresses the **symptom** by providing a numerically stable alternative to the Softmax function in the cross-entropy loss calculation.

**To create the ideal conditions to achieve grokking and ensure maximum numerical stability, using these two components together is highly recommended.**

> A PyTorch implementation of the `stablemax_cross_entropy` loss function is available as part of the **[Axolotl](https://github.com/OpenAccess-AI-Collective/axolotl)** fine-tuning framework. You can find the implementation and details here:
>
> **[https://github.com/cognitivecomputations/axolotl/blob/main/src/axolotl/integrations/stablemax/](https://github.com/cognitivecomputations/axolotl/blob/main/src/axolotl/integrations/stablemax/)**

## Performance: `torch.compile` and Mixed Precision

`DolphinFlow` is designed to be a modern, high-performance optimizer.

### `torch.compile` (Recommended)

For maximum performance, use `DolphinFlow` with `torch.compile`, the JIT compiler in PyTorch 2.0+. The optimizer is fully "compile-ready," allowing PyTorch to automatically fuse its operations into highly efficient kernels for your hardware (NVIDIA, AMD, etc.).

**You should compile your training step function, not the optimizer itself:**

```python
import torch
from dolphinflow import DolphinFlow

# Your optimizer and model
optimizer = DolphinFlow(...)

# Your training logic
def train_step(data, targets):
    optimizer.zero_grad()
    # model forward, loss, backward...
    optimizer.step()

# Compile the function for a massive speedup
compiled_train_step = torch.compile(train_step)

# Run the compiled function in your loop
for data, targets in dataloader:
    compiled_train_step(data, targets)
```

### Automatic Mixed Precision (`torch.amp`)

The standard 32-bit `DolphinFlow` is the perfect companion for `torch.amp`. For best results on modern GPUs (A100, H100), use `torch.autocast` with `dtype=torch.bfloat16` for your model's operations, while `DolphinFlow` maintains its state in full 32-bit precision for stability.

## `DolphinFlow` vs. `DolphinFlow8bit`

| Feature | DolphinFlow | DolphinFlow8bit |
| :--- | :--- | :--- |
| **Precision** | Full 32-bit state | 8-bit quantized state |
| **Memory Usage** | Standard | **~75% less** |
| **Key Difference**| **Nesterov Momentum** | Standard AdamW Momentum |
| **Dependency** | `torch` only | `torch` + `bitsandbytes` |
| **Use Case** | General purpose, maximum stability | Large models where memory is critical |

## Citation

If you use DolphinFlow in your research, please consider citing the software directly:

```bibtex
@misc{hartford2024dolphinflow,
  author = {Eric Hartford},
  title = {DolphinFlow: A Robust Orthogonalizing Optimizer for PyTorch},
  year = {2024},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/cognitivecomputations/dolphinflow-optimizer}}
}
```

## Acknowledgements and Inspirations

DolphinFlow is a pragmatic synthesis of several powerful ideas from the optimizer and deep learning literature. It stands on the shoulders of giants, and proper credit is essential.

*   The **vector-wise orthogonalization** (`ortho_mode="vector"`) is directly inspired by the `⊥Grad` concept introduced by **Lucas Prieto et al.** in their work on grokking. Their insight that preventing Naïve Loss Minimization can lead to faster generalization is a core motivation for this feature.
*   The **Newton-Schulz iteration** used in `ortho_mode="matrix"` adopts the highly effective quintic polynomial coefficients `(a=3.4445, b=-4.7750, c=2.0315)` discovered by **Keller Jordan et al.** for their Muon optimizer.
*   The general framework benefits from the robust design patterns of **AdamW** (decoupled weight decay) and **SGD** (Nesterov momentum).

We strongly encourage you to also read and cite these foundational works:

```bibtex
@misc{jordan2024muon,
  author       = {Keller Jordan and Yuchen Jin and Vlado Boza and You Jiacheng and
                  Franz Cesista and Laker Newhouse and Jeremy Bernstein},
  title        = {Muon: An optimizer for hidden layers in neural networks},
  year         = {2024},
  url          = {https://kellerjordan.github.io/posts/muon/}
}

@article{prieto2025grokking,
  title={Grokking at the Edge of Numerical Stability},
  author={Prieto, Lucas and Barsbey, Melih and Mediano, Pedro and Birdal, Tolga},
  year = {2025},
  eprint={2501.04697},
  archivePrefix={arXiv},
  primaryClass={cs.LG}
}
```

## License

MIT
