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

QServe: W4A8KV4 Quantization and System Co-design for Efficient LLM Serving (Zhong et al., NYU/CoatNet)

paper

Fields

id
sources/qlab-qserve-paper
URL
https://arxiv.org/abs/2405.04532
Publisher
arXiv
Kind
paper
Published
2024-05-07
Accessed
2026-10-03
notes
v3. QoQ = W4A8KV4; progressive quantization, SmoothAttention, compute-aware weight reordering, register-level parallelism.

Notes

v3. QoQ = W4A8KV4; progressive quantization, SmoothAttention, compute-aware weight reordering, register-level parallelism.

Cross-references

This record references no other records.