Allow q8_0 cache on the CPU for FlashMLA-2#265
Merged
Conversation
ikawrakow
pushed a commit
that referenced
this pull request
Mar 19, 2025
I broke it with PR #265. I was testing with a model where the wk_b and wk_v tensors were present, so didn't need to be computed, so didn't notice that the change I made to ggml_compute_forward_dup_q breaks that computation.
ikawrakow
added a commit
that referenced
this pull request
Mar 19, 2025
I broke it with PR #265. I was testing with a model where the wk_b and wk_v tensors were present, so didn't need to be computed, so didn't notice that the change I made to ggml_compute_forward_dup_q breaks that computation. Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Somehow I had the concept that
Q8_0KV cache is working for CPU-only inference with FlashMLA-2. Indeed it is for prompt processing, but not for TG (two different paths are taken). Clearly too many options as I'm getting confused myself. Anyhow, this PR adds the missingQ8_0 -> Q8_0contiguous transpose operation, so now we can useQ8_0KV cache with FlashMLA-2 also on the CPU.