One of the focus areas at Together Research is new architectures for long context, improved training, and inference performance over the Transformer architecture. Spinning out of a research program from our team and academic collaborators, with roots in signal processing-inspired sequence models, we are excited to introduce the StripedHyena models. StripedHyena is the first alternative model competitive with the best open-source Transformers of similar sizes in short and long-context evaluations.
StripedHyena-Nous-7B (SH-N 7B) is our chat model for this release, and was developed with our collaborators at Nous Research.
StripedHyena is a hybrid architecture composed of multi-head, grouped-query attention and gated convolutions arranged in Hyena blocks, different from traditional decoder-only Transformers.
Costant memory decoding in Hyena blocks via representation of convolutions as state-space models (modal or canonical form), or as truncated filters.
Low latency, faster decoding and higher throughput than Transformers.
Improvement to training and inference-optimal scaling laws, compared to optimized Transformer architectures such as Llama-2.
Trained on sequences of up to 32k, allowing it to process longer prompts.
To use StripedHyena outside of the playground, you will need to install custom kernels. Please follow the instructions from the standalone repository.
StripedHyena is a mixed precision model. Make sure to keep your poles and residues in float32 precision, especially for longer prompts or training.
Cite
If you have found the pretrained models or architecture useful for you research or application, consider citing:
@software{stripedhyena,
title = {{StripedHyena: Moving Beyond Transformers with Hybrid Signal Processing Models}},
author = { Poli, Michael and Wang, Jue and Massaroli, Stefano and Quesnelle, Jeffrey and Carlow, Ryan and Nguyen, Eric and Thomas, Armin},
month = 12,
year = 2023,
url = { https://github.com/togethercomputer/stripedhyena },
doi = { 10.57967/hf/1595 },
}
