inference/research
verifiedhigh · 0.88updated 2026-10-03

H2O: Heavy-Hitter Oracle for Efficient Generative Inference of Large Language Models (Zhang et al., MBZUAI)

paper

Fields

id
sources/qlab-h2o-paper
URL
https://arxiv.org/abs/2306.14048
Publisher
arXiv
Kind
paper
Published
2023-06-24
Accessed
2026-10-03
notes
v3. Eviction policy retaining heavy-hitter plus recent tokens. Explicitly NOT a KV quantization method - recorded here only so it is not confused with one.

Notes

v3. Eviction policy retaining heavy-hitter plus recent tokens. Explicitly NOT a KV quantization method - recorded here only so it is not confused with one.

Cross-references

This record references no other records.