Improve language tag (#1)
Browse files- Improve language tag (0da932bd1be64f91b1ea1ea027ca0b3cae42227b)
Co-authored-by: Loïck BOURDOIS <[email protected]>
README.md
CHANGED
|
@@ -1,92 +1,104 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
datasets:
|
| 4 |
-
- allenai/c4
|
| 5 |
-
language:
|
| 6 |
-
-
|
| 7 |
-
|
| 8 |
-
-
|
| 9 |
-
-
|
| 10 |
-
|
| 11 |
-
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
| 32 |
-
|
| 33 |
-
|
| 34 |
-
<
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
|
| 50 |
-
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
```python
|
| 56 |
-
|
| 57 |
-
|
| 58 |
-
|
| 59 |
-
|
| 60 |
-
|
| 61 |
-
|
| 62 |
-
|
| 63 |
-
|
| 64 |
-
|
| 65 |
-
|
| 66 |
-
|
| 67 |
-
|
| 68 |
-
|
| 69 |
-
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
|
| 73 |
-
|
| 74 |
-
|
| 75 |
-
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
|
| 81 |
-
|
| 82 |
-
|
| 83 |
-
|
| 84 |
-
|
| 85 |
-
|
| 86 |
-
|
| 87 |
-
|
| 88 |
-
|
| 89 |
-
|
| 90 |
-
|
| 91 |
-
|
| 92 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
datasets:
|
| 4 |
+
- allenai/c4
|
| 5 |
+
language:
|
| 6 |
+
- zho
|
| 7 |
+
- eng
|
| 8 |
+
- fra
|
| 9 |
+
- spa
|
| 10 |
+
- por
|
| 11 |
+
- deu
|
| 12 |
+
- ita
|
| 13 |
+
- rus
|
| 14 |
+
- jpn
|
| 15 |
+
- kor
|
| 16 |
+
- vie
|
| 17 |
+
- tha
|
| 18 |
+
- ara
|
| 19 |
+
metrics:
|
| 20 |
+
- perplexity
|
| 21 |
+
- accuracy
|
| 22 |
+
base_model:
|
| 23 |
+
- Qwen/Qwen2.5-7B
|
| 24 |
+
pipeline_tag: text-generation
|
| 25 |
+
library_name: transformers
|
| 26 |
+
---
|
| 27 |
+
<div align="center">
|
| 28 |
+
<img width="30%" alt="logo" src="https://imgur.com/A0MCHPq.png">
|
| 29 |
+
</div>
|
| 30 |
+
|
| 31 |
+
<div align="center">
|
| 32 |
+
<a href="https://github.com/merantix-momentum/acip"><img src="https://img.shields.io/badge/GitHub-%23121011.svg?logo=github&logoColor=white.svg" alt="github" style="display: inline-block; vertical-align: middle;"></a>
|
| 33 |
+
<a href="https://arxiv.org/abs/2502.01717"><img src="https://img.shields.io/badge/arXiv-2502.01717-b31b1b.svg" alt="arxiv" style="display: inline-block; vertical-align: middle;"></a>
|
| 34 |
+
<a href="https://acip.merantix-momentum.com/"><img alt="website" src="https://img.shields.io/website/https/acip.merantix-momentum.com.svg?down_color=red&down_message=offline&up_message=online" style="display: inline-block; vertical-align: middle;"></a>
|
| 35 |
+
</div>
|
| 36 |
+
|
| 37 |
+
<h2 align="center">
|
| 38 |
+
<p> [
|
| 39 |
+
<a href="https://github.com/merantix-momentum/acip">🤖 GitHub</a> |
|
| 40 |
+
<a href="https://arxiv.org/abs/2502.01717">📄 Paper</a> |
|
| 41 |
+
<a href="https://acip.merantix-momentum.com/">🌐 Website</a>
|
| 42 |
+
]
|
| 43 |
+
</p>
|
| 44 |
+
</h2>
|
| 45 |
+
|
| 46 |
+
<h1 align="center">
|
| 47 |
+
<p>ACIP applied to Qwen/Qwen2.5-7B</p>
|
| 48 |
+
</h1>
|
| 49 |
+
|
| 50 |
+
This model repository is part of the ACIP Project and provides a compressible version of [`Qwen/Qwen2.5-7B`](https://huggingface.co/Qwen/Qwen2.5-7B). For more details, please visit our [code repo](https://github.com/merantix-momentum/acip).
|
| 51 |
+
|
| 52 |
+
# Quick Start
|
| 53 |
+
|
| 54 |
+
Just load the ACIP model via `from_pretrained`:
|
| 55 |
+
```python
|
| 56 |
+
from transformers import AutoModel
|
| 57 |
+
|
| 58 |
+
model = AutoModel.from_pretrained("MerantixMomentum/acip_qwen25_7b", trust_remote_code=True)
|
| 59 |
+
```
|
| 60 |
+
This will download and create a fully parameterized ACIP model that can be pruned to any compression ratio you wish.
|
| 61 |
+
For example,
|
| 62 |
+
```python
|
| 63 |
+
model.prune_model_by_score(compression_ratio=0.4)
|
| 64 |
+
```
|
| 65 |
+
will prune `model` to 40% if its original size measured in number of parameters, i.e., 60% compression rate.
|
| 66 |
+
A unique feature of ACIP is that this operation is revertible in the sense that you can rerun `model.prune_model_by_score` as often as you like to evaluate your model at different sizes. Finally, you can "commit" to a certain ratio and run
|
| 67 |
+
```python
|
| 68 |
+
model.compress()
|
| 69 |
+
```
|
| 70 |
+
which will discard all pruned mask values of compressible linear layers.
|
| 71 |
+
Now the model is actually compressed and you should observe a significant decrease of memory usage (this step is not revertible without reloading the ACIP model).
|
| 72 |
+
If you like, you can also run
|
| 73 |
+
```python
|
| 74 |
+
model.quantize()
|
| 75 |
+
```
|
| 76 |
+
to save even more memory (we have only tested 4bit quantization with `bitsandbytes`, but you could also customize this).
|
| 77 |
+
|
| 78 |
+
**🚀 That's it! You can now use your compressed model for inference or fine-tuning as any other Causal Language Model from 🤗 transformers.**
|
| 79 |
+
|
| 80 |
+
**Note**: The parameter `compression_ratio` ranges from 1.0 to 0.0, indicating the model size after compression. For example, 0.4 means that the model has only 40% of the original number of parameters and 1.0 means no compression at all.
|
| 81 |
+
|
| 82 |
+
# Dependencies
|
| 83 |
+
|
| 84 |
+
To run an ACIP model from our hub, you only need minimal dependencies, namely `torch`, `transformers`, `peft`, and optionally, `bitsandbytes` in case you want to quantize your model.
|
| 85 |
+
See [requirements.txt](requirements.txt) for pip-installable dependencies with exact version pins (newer version should work as well).
|
| 86 |
+
|
| 87 |
+
# License
|
| 88 |
+
|
| 89 |
+
This model is released under the apache-2.0 license.
|
| 90 |
+
|
| 91 |
+
# Citation
|
| 92 |
+
|
| 93 |
+
When using or referring to this model, please cite our [paper](https://arxiv.org/abs/2502.01717):
|
| 94 |
+
```bibtex
|
| 95 |
+
@article{mxm2025acip,
|
| 96 |
+
title={Choose Your Model Size: Any Compression by a Single Gradient Descent},
|
| 97 |
+
author={M. Genzel, P. Putzky, P. Zhao, S. Schulze, M. Mollenhauer, R. Seidel, S. Dietzel, T. Wollmann},
|
| 98 |
+
year={2025},
|
| 99 |
+
journal={Preprint arXiv:2502.01717}
|
| 100 |
+
}
|
| 101 |
+
```
|
| 102 |
+
|
| 103 |
+
|
| 104 |
+
|