English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Tools annual

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools monthly

Global: Zendy Plus monthly

Presents the access of premium content as premium feature

Premium Content

Global: Zendy Plus annual

Global: Zendy Free Plan

Large Language Models (LLMs) have demonstrated impressive performance on arange of Natural Language Processing (NLP) tasks. Unfortunately, the immenseamount of computations and memory accesses required for LLM training makes themprohibitively expensive in terms of hardware cost, and thus challenging todeploy in use cases such as on-device learning. In this paper, motivated by theobservation that LLM training is memory-bound, we propose a novel dynamicquantization strategy, termed Dynamic Stashing Quantization (DSQ), that puts aspecial focus on reducing the memory operations, but also enjoys the otherbenefits of low precision training, such as the reduced arithmetic cost. Weconduct a thorough study on two translation tasks (trained-from-scratch) andthree classification tasks (fine-tuning). DSQ reduces the amount of arithmeticoperations by $20.95\times$ and the number of DRAM operations by $2.55\times$on IWSLT17 compared to the standard 16-bit fixed-point, which is widely used inon-device learning.

Dynamic Stashing Quantization for Efficient Transformer Training