]> git.djapps.eu Git - pkg/ggml/sources/llama.cpp/commit
llama : load MTP tensors only if they are really used (#26296)
authorfairydreaming <redacted>
Fri, 31 Jul 2026 12:57:02 +0000 (14:57 +0200)
committerGitHub <redacted>
Fri, 31 Jul 2026 12:57:02 +0000 (14:57 +0200)
commit82dbc4f017a7b005f993ac2e7af9c048ad686c04
treee6a7f01c7e397557178a1bb21b1383659cd0095c
parent6f3c0a790be2c8599663a98f03b95fb3eae767f8
llama : load MTP tensors only if they are really used (#26296)

* llama : load MTP tensors only if they are really used

* llama : skip loading MTP (if not used) in remaining models that support MTP

---------

Co-authored-by: Stanisław Szymczyk <redacted>
13 files changed:
common/common.cpp
include/llama.h
src/llama-model-loader.cpp
src/llama-model-loader.h
src/llama-model.cpp
src/llama-quant.cpp
src/llama.cpp
src/models/cohere2moe.cpp
src/models/glm-dsa.cpp
src/models/hy-v3.cpp
src/models/qwen35.cpp
src/models/qwen35moe.cpp
src/models/step35.cpp