The authors first analyzed how the KV cache evolves across latent frames in autoregressive video diffusion models.
TempoKV compresses video generation KV caches by exploiting key-value predictability asymmetry
The training-free framework predicts keys from temporal history and values from aligned keys, quantizing only residuals to reach up to 6.71x compression.
Chinese Tech
Jiahui Hu · Weile Mo · Peiliang Cai · Jiacheng Liu · Jinkui Ren · Xiantao Zhang · +2 more
Shanghai Jiao Tong University · South China University of Technology · Alibaba Group
Research Digest··2 min read
The authors propose TempoKV, a training-free quantization framework that compresses the KV cache, the growing store of attention keys and values, in autoregressive video diffusion models.
Why this paper
From Alibaba Group and 2 others
In one line
TempoKV predicts keys from history and values from current keys to compress KV cache up to 6.71x for video generation.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§