What's the difference between LMCache and KVBM for cpu offloading?

[janetat]

Both LMCache and KVBM provides KV cache cpu offloading.
So What's the diff?

[bai030805]

Same confuse :)

[luckyq]

Generally, I do not think there is a fundamental difference.

Both KVBM and LMCache follow vLLM’s one-sided contract. You can certainly implement your own KV cache management beyond LMCache or KVBM, but if you want it to remain compatible with vLLM, you still need to follow that contract. My sense is that vLLM would likely be open to providing a richer contract if there were a strong enough need.

That said, this is just my current understanding.