There are some transformation approaches to re-use the kv cache across inferences, but none are in wide use due to accuracy concerns following the transformation.
My understanding was that what the KV cache stores is nothing else than the "activations" of the W_k and W_v matrices of an attention module for a given input sequence.
So I don't quite understand how this is supposed to work:
> Let a publisher precompute a document's KV cache, and let every other agent buy the right to load it and skip prefill.
Should a publisher precompute the cache for every popular model that is out there?
"People are asking the same questions and an answer is generated every time, what if we could like cache the questions and their answers..."
Sounds like someone was using chatgpt to understand how chatgpt works and then asked it to generate a paper based on his proposal to improve it.
You basically have two agents look at the same cache under different views. Say agent_0 gets [a_1, a_0] and agent_1 gets [a_0, a_1]. They also write to this cache concurrently while decoding. To solve positional embedding inconsistencies they rotate the query projections for each block (a_0 and a_1) separately.
The computations you get that way do not exactly match the setup where you would naively prefill on every step, but are close enough.
Same trick could be used for the setup discussed here, I guess: prefill the document cache separately (p), prepend the system prompt (s) and get a cache view [s, p] from which you can then decode.