feat: First pass at llama_kv_cache_hybrid #13276
Conversation
This implementation covers both `llama_memory_i` and `llama_kv_cache` interfaces, but they could very well not be correct. Branch: HybridCache Signed-off-by: Gabe Goodhart <[email protected]>
| // TODO: Will it cause problems if some caches are able to remove the seq | ||
| // but others aren't? |
There was a problem hiding this comment.
Yes it will cause problems if this breaks the coherency between caches. (e.g. part of a sequence is removed in one cache but not the other).
This is what I was referring to in #12799 (comment) when I wrote:
The hardest part will be handling errors and properly keeping coherency between the different types of caches (because they don't necessarily roll-back states in the same way).
I think the seq_rm API might fundamentally be too specific to self-attention KV cache. Recurrent models can't rollback their state, because intermediate states are not kept since keeping them for all tokens would take too much space. (when seq_rm returns false, it means the states have to be re-calculated from scratch for the affected sequence (at least that was the intention in #5328))
Ideally, if there was some API to create snapshots and rollback to them, the implementation would be simpler for recurrent models (and for hybrid models by extension). (technically, sequences (with seq_id) already kind of do this (and are copy-on-write), but snapshots within sequences might be more convenient to manage in user code, since managing which state is the latest per sequence could be done transparently)
But that would also mean having to manage the lifetime of explicit state snapshots (in examples/server/server.cpp among others) instead of directly dealing with ranges of token positions (and might make things like largest-common-prefix context caching harder to handle). I've previously shared some ideas about state snapshots/checkpoints in #7531 (comment) (although the first half of the comment is about session restore as in state_read).
| // If any of the caches are recurrent, require simple split | ||
| return llama_sbatch(batch, m_hparams.n_embd, m_has_recurrent, logits_all); |
There was a problem hiding this comment.
Simple split should not be used with recurrent models, they expect equal split.
See #7531 (comment) which illustrates the splits
| // If any of the caches are recurrent, require simple split | |
| return llama_sbatch(batch, m_hparams.n_embd, m_has_recurrent, logits_all); | |
| // If any of the caches are recurrent, require non-simple split | |
| return llama_sbatch(batch, m_hparams.n_embd, !m_has_recurrent, logits_all); |
| if (m_has_recurrent) { | ||
| return sbatch.split_simple(n_ubatch); | ||
| } |
There was a problem hiding this comment.
This will not work, recurrent models expect split_equal to be used.
| // TODO: Is this correct? | ||
| // If any children can shift, return true | ||
| for (const auto & cache : m_children) { | ||
| if (cache->get_can_shift()) { | ||
| return true; | ||
| } | ||
| } |
There was a problem hiding this comment.
Maybe this should be if all children can shift, then return true.
But as you've noticed elsewhere, can_shift should technically always be true for all currently-implemented cache types, so I don't know if that part of the API will stay anyway.
Description
This implementation covers both
llama_memory_iandllama_kv_cacheinterfaces, but they could very well not be correct.Discussion
I'm putting this up for discussion even though it doesn't have much value as standalone. My ultimate goal is support for the just-released granite 4 which is a combination of
mamba2andgranitemoesharedlayers. I opened #13275 to track the full scope of model architecture changes.