GoldFinch: High Performance RWKV/Transformer Hybrid with Linear Pre-Fill and Extreme KV-Cache Compression
Abstract
We introduce GoldFinch, a hybrid Linear Attention/Transformer sequence model that uses a new technique to efficiently generate a highly compressed and reusable KV-Cache in linear time and space with respect to sequence length. GoldFinch stacks our new GOLD transformer on top of an enhanced version of the Finch (RWKV-6) architecture. We train up to 1.5B parameter class models of the Finch, Llama, and GoldFinch architectures, and find dramatically improved modeling performance relative to both Finch and Llama. Our cache size savings increase linearly with model layer count, ranging from 756-2550 times smaller than the traditional transformer cache for common sizes, enabling inference of extremely large context lengths even on limited hardware. Although autoregressive generation has O(n) time complexity per token because of attention, pre-fill computation of the entire initial cache state for a submitted context costs only O(1) time per token due to the use of a recurrent neural network (RNN) to generate this cache. We release our trained weights and training code under the Apache 2.0 license for community use.
Community
Hi @Fareso congrats on this work!
Would you be able to link the model to this paper page? I opened a PR here: https://huggingface.co/recursal/GoldFinch-paper/discussions/1.
Also, note that we recommend pushing each checkpoint to a separate model repository, so that things like download stats work. Read more here:
- uploading models: https://huggingface.co/docs/hub/models-uploading
- making download stats work: https://huggingface.co/docs/hub/models-download-stats
Let me know if you need any help!
Cheers,
Niels
@librarian-bot recommend
This is an automated message from the Librarian Bot. I found the following papers similar to this paper.
The following papers were recommended by the Semantic Scholar API
- Reducing Transformer Key-Value Cache Size with Cross-Layer Attention (2024)
- LoCoCo: Dropping In Convolutions for Long Context Compression (2024)
- MLKV: Multi-Layer Key-Value Heads for Memory Efficient Transformer Decoding (2024)
- Effectively Compress KV Heads for LLM (2024)
- Zamba: A Compact 7B SSM Hybrid Model (2024)
Please give a thumbs up to this comment if you found it helpful!
If you want recommendations for any Paper on Hugging Face checkout this Space
You can directly ask Librarian Bot for paper recommendations by tagging it in a comment:
@librarian-bot
recommend
Interesting! Eagle/Finch look awesome in this paper: https://huggingface.co/papers/2404.05892.
Love to hear your thoughts on the reason behind.
Models citing this paper 1
Datasets citing this paper 0
No dataset linking this paper
Spaces citing this paper 0
No Space linking this paper