Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/google-deepmind/recurrentgemma

Open weights language model from Google DeepMind, based on Griffin.
https://github.com/google-deepmind/recurrentgemma

Last synced: 3 months ago
JSON representation

Open weights language model from Google DeepMind, based on Griffin.

Awesome Lists containing this project

README 

        
# RecurrentGemma



RecurrentGemma is a family of open-weights Language Models by [Google DeepMind](https://deepmind.google/), based on the novel [Griffin architecture](https://arxiv.org/abs/2402.19427). This architecture achieves fast inference when generating long sequences by replacing global attention with a mixture of local attention and linear recurrences.



This repository contains the model implementation and examples for sampling and fine-tuning. We recommend most users adopt the [Flax](https://github.com/google/flax) implementation, which is highly optimized. We also provide an un-optimized [PyTorch](https://github.com/pytorch/pytorch) implementation for reference.



### Learn more about RecurrentGemma



-   The [RecurrentGemma technical report](https://storage.googleapis.com/deepmind-media/gemma/recurrentgemma-report.pdf) gives specific details on the training and evaluation of RecurrentGemma.

-   The [Griffin paper](https://arxiv.org/abs/2402.19427) describes the underlying model architecture.



## Quick start



### Installation



#### Using Poetry

RecurrentGemma uses [Poetry](https://python-poetry.org/docs/) for dependency

management.



To install dependencies for the full project:

* Checkout the code.

* `poetry install -E full` to create a virtual environment with all dependencies.

* `poetry shell` to activate the created virtual environment.



If you only need to install a subset of dependencies use one of the alternative

library-specific commands below.



#### Using pip

If you want to use `pip` instead of Poetry, 

then create a virtual environment (run `python -m venv recurrentgemma-demo` and `. recurrentgemma-demo/bin/activate`) and:



* Checkout the code.

* `pip install .[full]`



#### Installing library-specific packages



##### JAX

To install dependencies only for the JAX pathway use:

`poetry install -E jax` or (`pip install .[jax]`).



##### PyTorch

To install dependencies only for the PyTorch pathway use:

`poetry install -E torch` (or `pip install .[torch]`).



##### Tests

To install dependencies required for running unit tests use:

`poetry install -E test` (or `pip install .[test]`)



### Downloading the models



The model checkpoints are available through Kaggle at

http://kaggle.com/models/google/recurrentgemma.

Select either the **Flax** or **PyTorch** model variations, click the ⤓ button

to download the model archive, then extract the contents to a local directory.



In both cases, the archive contains both the model weights and

the tokenizer.



### Running the unit tests



To run the tests, install the optional `[test]` dependencies (e.g. using `pip install .[test]`) from the root of the source tree, then:



```

pytest .

```



## Examples



To run the example sampling script, pass the paths to the weights directory and tokenizer:



```

python examples/sampling_jax.py \

  --path_checkpoint=/path/to/archive/contents/2b/ \

  --path_tokenizer=/path/to/archive/contents/tokenizer.model

```



### Colab notebook tutorials



-   [`colabs/sampling_tutorial_jax.ipynb`](https://colab.sandbox.google.com/github/google-deepmind/recurrentgemma/blob/main/colabs/sampling_tutorial_jax.ipynb)

    contains a [Colab](http://colab.google) notebook with a sampling example using JAX.



-   [`colabs/sampling_tutorial_pytorch.ipynb`](https://colab.sandbox.google.com/github/google-deepmind/recurrentgemma/blob/main/colabs/sampling_tutorial_pytorch.ipynb)

    contains a [Colab](http://colab.google) notebook with a sampling example using PyTorch.



-   [`colabs/fine_tuning_tutorial_jax.ipynb`](https://colab.sandbox.google.com/github/google-deepmind/recurrentgemma/blob/main/colabs/fine_tuning_tutorial_jax.ipynb)

    contains a [Colab](http://colab.google) with a basic tutorial on how to

    fine-tune RecurrentGemma for a task, such as English to French translation, using JAX.



To run these notebooks you will need to have a Kaggle account and first read and accept

the Gemma license terms and conditions from the [RecurrentGemma page](http://kaggle.com/models/google/recurrentgemma).

After this you can run the notebooks, which will automatically download the weights and tokenizer from there.



Currently different notebooks are supported under the following hardware:



| Hardware            | T4  | P100 | V100 | A100 | TPUv2 | TPUv3+ |

|---------------------|:---:|:----:|:----:|:----:|:-----:|:------:|

| Sampling in Jax     | ✅  | ✅   | ✅   | ✅   | ✅    | ✅    |

| Sampling in PyTorch | ✅  | ✅   | ✅   | ✅   | ✅    | ✅    |

| Finetuning in Jax   | ✅  | ✅   | ✅   | ✅   | ❌    | ✅    |



## System Requirements



RecurrentGemma code can run on CPU, GPU or TPU.

The code has been optimized for running on TPU using the Flax implementation,

which contains a low level [Pallas](https://jax.readthedocs.io/en/latest/pallas/index.html) kernel to perform the linear scan in the recurrent layers.



## Contributing



We are open to bug reports and issues. Please see

[CONTRIBUTING.md](CONTRIBUTING.md) for details on PRs.



## License



Copyright 2024 DeepMind Technologies Limited



This code is 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.



## Disclaimer



This is not an official Google product.