Unofficial Implementation of "Mamba: Linear-Time Sequence Modeling with Selective State Spaces" in JAX.

⚠️ This is very much a work-in-progress implementation. Expect numerical mismatches, slower speeds, bad code, and general wrongness herein. ⚠️

As the plan is to eventually write custom Pallas kernels for the Mamba recurrence scan, we need to install requirements that work with Pallas.

Unfortunately, Pallas is currently quite hard to install (see this issue) and the required options can't be fully specified in a requirements.txt file. So, to setup the environment for this repository, take the following steps:

1. Create a Python 3.9 or 3.10 virtual environment.
2. Run install-requirements.txt and ensure none of the commands fail.

Such a kernel does not exist yet, and it is not clear how it would be implemented. However, I optimistically pin the versions for now.

The script sample.py is the main entry point to sample from a pretrained Mamba model:

usage: sample.py [-h] [--prompt PROMPT] [--model MODEL] [--bf16] [--gen_len GEN_LEN]
                 [--temperature TEMPERATURE] [--seed SEED] [--seed_iters SEED_ITERS]
                 [--scan]

options:
  -h, --help            show this help message and exit
  --prompt PROMPT       Starting prompt for generation. (default: Aloha, World! )
  --model MODEL         Model repo id as on Huggingface Hub. (default: state-
                        spaces/mamba-2.8b)
  --bf16                Use bfloat16 for inference (default: False)
  --gen_len GEN_LEN     Length of generated sequence. (default: 1024)
  --temperature TEMPERATURE
                        Sampling temperature. (default: 1.0)
  --seed SEED           Random seed for PRNG initialisation. (default: 0)
  --seed_iters SEED_ITERS
                        Number of seeds to generate, starting from --seed. (default: 1)
  --scan                Use jax.lax.scan version of generate loop. (default: False)

The components of the full Mamba architecture can be imported as follows:

• An interface with the S6 (S4 with selective scan) model can be imported at the path mamba_jax.kernels.mamba_ssm. This is a purely functional implementation of Algorithm 2 in the paper which is agnostic of the neural network API you use. Currently, this just dispatches to a pure JAX implementation, though the idea is you will be able to dispatch to an optimised Pallas kernel via the mode argument in the future.
• Equinox Mamba language model and sub-components of it can be found in mamba_jax.modelling.equinox as MambaBlock, ResidualBlock, MambaModel, and MambaLLM.
• PRs for other neural network APIs (Flax, NNX) welcome.

• Make this all pip installable.
• Testing to 100% verify parity with CUDA reference.
• Add efficient training code in pure JAX.
• Add efficient custom kernels for work-efficient associative scan, implemented in Pallas.
• Try to reproduce some training results from scratch.
• Complex number mode

This implementation was based off a mix of:

• Original implementation in PyTorch and CUDA
• Minimal Mamba implementation in PyTorch

A lot of understanding of how S4 models work was derived from:

• Albert Gu's amazing thesis on the topic
• The Annotated S4 by Sasha Rush

And a lot of understanding on the associative scan recurrent form was derived from:

• Appendix H of the S5 paper

Mamba: Linear-Time Sequence Modeling with Selective State Spaces

Albert Gu, Tri Dao

@misc{gu2023mamba,
      title={Mamba: Linear-Time Sequence Modeling with Selective State Spaces},
      author={Albert Gu and Tri Dao},
      year={2023},
      eprint={2312.00752},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}