Tiny deterministic model predicts the K/V cache, prediction is compared with reality, delta is stored in vram.
The other way round then just predicts the values again, applies the delta, and you have the full correct value while just storing the delta
And this works because you're never looking at the whole k/v cache but always just a slice. So you just need a memory buffer of the size of the slice
___
If this works out and I've understood correctly, that _I think_ would mean that a 24GB RTX 4090 could fit 256k q8 context next to Qwen3.6-27B at IQ4_NL.
Or, alternatively, something like 208k context (matching claude api limits of 200k in some plans) with a slightly larger quant like UD-Q4_K_XL.
That would be massive. Especially since the thing has so much compute to spare.
Though, all depending on the size of that predictor model I guess?
porridgeraisin•1h ago
I am yet to do a "deep dive" into the results, but what a well written article. An LLM could _never_ write so crisply.
mirekrusin•1h ago
If “speculative” approach works so well in different contexts why not make it first class and use everywhere, possibly recursively?
0-_-0•43m ago
You can use the original model to compress the kv cache and get ∞x compression, since the prediction is perfect. The cost is time, and I don't see how this could be worth it.
zozbot234•41m ago
The problem with this approach is that even recomputing a "draft" of the KV cache is still quadratic in context length. Maybe you can get some constant savings by always recomputing the earliest tokens, but it's not a good tradeoff as context sizes grow.
hypfer•1h ago
Tiny deterministic model predicts the K/V cache, prediction is compared with reality, delta is stored in vram. The other way round then just predicts the values again, applies the delta, and you have the full correct value while just storing the delta
And this works because you're never looking at the whole k/v cache but always just a slice. So you just need a memory buffer of the size of the slice
___
If this works out and I've understood correctly, that _I think_ would mean that a 24GB RTX 4090 could fit 256k q8 context next to Qwen3.6-27B at IQ4_NL.
Or, alternatively, something like 208k context (matching claude api limits of 200k in some plans) with a slightly larger quant like UD-Q4_K_XL.
That would be massive. Especially since the thing has so much compute to spare.
Though, all depending on the size of that predictor model I guess?