The authors studied cache-to-cache communication in multi-agent LLM systems, focusing on the scenario where both models are frozen and no fine-tuning occurs.
KV cache substitution between frozen LLMs fails; augmentation succeeds
The authors demonstrate that frozen transformers can only use caches within a narrow 'tolerable set' and that adding the sender's cache as extra attendable context outperforms substitution.
Big Tech
Adam Deng · Shicheng Liu · Yada Zhu
MIT ORC · MIT CSAIL · IBM Research
Research Digest··3 min read
Thread:Cross-Model KV Cache Reuse
The authors tested eight methods for substituting a sender's KV cache into a frozen reader LLM in place of its own cache.
Why this paper
From IBM Research and 2 others
In one line
Frozen transformers cannot use another model's KV cache as a substitute; caches must be added as extra context.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (2 noted)
- ·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.
§