graph : create V as a view of K in the k_iswa build_attn - #27392
Merged
Conversation
build_attn with the llm_graph_input_attn_k_iswa input was using the cached K tensor itself as V. Create V as a view of K (the first v_cur->ne[0] elements of each row), like the other K-only build_attn overloads. The deepseek4 MTP call site now passes the kv tensor as v_cur. Assisted-by: pi:llama.cpp/Qwen3.8-27B
am17an
approved these changes
Aug 19, 2026
CISC
approved these changes
Aug 19, 2026
Member
Author
|
@am17an A quick look in the |
therealkenc
pushed a commit
to therealkenc/llama.cpp
that referenced
this pull request
Aug 24, 2026
) build_attn with the llm_graph_input_attn_k_iswa input was using the cached K tensor itself as V. Create V as a view of K (the first v_cur->ne[0] elements of each row), like the other K-only build_attn overloads. The deepseek4 MTP call site now passes the kv tensor as v_cur. Assisted-by: pi:llama.cpp/Qwen3.8-27B
ravel7524
pushed a commit
to ravel7524/llama.cpp
that referenced
this pull request
Aug 30, 2026
) build_attn with the llm_graph_input_attn_k_iswa input was using the cached K tensor itself as V. Create V as a view of K (the first v_cur->ne[0] elements of each row), like the other K-only build_attn overloads. The deepseek4 MTP call site now passes the kv tensor as v_cur. Assisted-by: pi:llama.cpp/Qwen3.8-27B
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
3 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Overview
In
build_attnfor thellm_graph_input_attn_k_iswainput (MLA-style attention where the cached K is used as V), V was the cached K tensor itself. It is now created as a view of K (the firstv_cur->ne[0]elements of each row), like the other K-onlybuild_attnoverloads (llm_graph_input_attn_k,llm_graph_input_attn_k_dsa).All current callers (deepseek4 MTP block, dflash DSpark decoder) have equal K/V head sizes, so the view is full-size and the attention results are unchanged — V simply becomes a proper view of K, which lets backends detect the V-is-view-of-K pattern.
The deepseek4 MTP call site now passes the
kvtensor asv_cur(it passednullptrbefore).Additional information
Related: #27390 (Metal FA: skip the redundant V dequant when V is a view of K)
Requirements