Tag
This paper proposes a method for cross-model memory transfer through target-side reader adaptation, using Engram hash memory and a lightweight reader, achieving 38.8 on QA tasks, and applicable to Agent memory updates and audits. The limitation is that it was only tested up to 9B models, with scaling laws unknown.
A technical blog post that explores sorting, hashing, and sketching algorithms on a dataset of 370,103 English words, measuring time and memory costs, with a focus on practical implementations like binary search, quicksort, and HyperLogLog.
Introduces BinaryPC, a training-free hashing-based sparse attention method for long-context LLMs that uses binary principal components to construct hash codes, preserving accuracy while improving decoding throughput by 3.56x over FlashAttention.
This article explains how implementing a Dense Arena Interner can drastically improve compiler performance by converting strings and structures into dense integers, enabling O(1) comparisons after an upfront hashing cost during lexing.
The Hugging Face CLI's `hf upload` command has been fully rewritten with single-pass hashing, multi-commits, and resumable uploads, available in version 1.20.0 for improved speed and cleanliness.
Bab is a family of hashing functions designed for peer-to-peer networks, enabling verifiable partial data. It was funded by the NLnet Foundation and the European Commission's Next Generation Internet programme.