llama.cpp
model: M3: Move MSA into a new memory implementation
#26338

Merged

model: M3: Move MSA into a new memory implementation #26338

ggerganov merged 5 commits into ggml-org:master from timkhronos:MSA-block-selection-kv-cell

timkhronos requested a review from

CISC 22 days ago

timkhronos requested a review from

ggerganov 22 days ago

github-actions added model

Move MSA logic from llama-kv-cache into llama-kv-cache-msa

f7e27bda

timkhronos force pushed from 938640f6 to f7e27bda 22 days ago

timkhronos changed the title ~~model: M3: Decouple MSA block selection from kv cell layout~~ model: M3: Decouple MSA block selection from kv cell layout and move MSA into a new llama_kv_cache_msa memory implementation 21 days ago

timkhronos changed the title ~~model: M3: Decouple MSA block selection from kv cell layout and move MSA into a new llama_kv_cache_msa memory implementation~~ model: M3: Move MSA into a new llama_kv_cache_msa memory implementation 21 days ago

timkhronos changed the title ~~model: M3: Move MSA into a new llama_kv_cache_msa memory implementation~~ model: M3: Move MSA into a new memory implementation 21 days ago

ggerganov assigned

ggerganov 21 days ago

Merge branch 'ggml-org:master' into MSA-block-selection-kv-cell

ad246695

Merge branch 'master' into MSA-block-selection-kv-cell

5be673d5

cont : minor

455db726

ggerganov approved these changes on 2026-08-03

ggerganov added merge ready

cont : ws fix

2248ec7d

ggerganov merged 67d5978b into master 18 days ago

Reviewers

ggerganov

CISC

Assignees

ggerganov

Labels

model merge ready

Milestone

No milestone

llama.cpp model: M3: Move MSA into a new memory implementation #26338 Merged

model: M3: Move MSA into a new memory implementation #26338

llama.cpp
model: M3: Move MSA into a new memory implementation
#26338

Merged