PR #23398 llama : add Gemma4 MTP - SemanticDiff

Commits

llama: Gemma 4 MTP

am17an committed 61 days ago
fix multi-seq

am17an committed 61 days ago
add assert that draft + shared kv should be on same device

am17an committed 61 days ago
add Q rot when cache is quantized

am17an committed 61 days ago
add temp hack to not use fit with gemma4, rm later

am17an committed 61 days ago
add exception in test-llama-archs

am17an committed 61 days ago
move assistant to separate file

am17an committed 61 days ago
add unified assistant

am17an committed 60 days ago
Merge branch 'master' into pr/23398

ggerganov committed 60 days ago
cont : adjust to hparams changes

ggerganov committed 60 days ago
cont : avoid computations on the CPU

ggerganov committed 60 days ago
Merge branch 'master' into pr/23398

ggerganov committed 59 days ago
Merge remote-tracking branch 'origin/master' into pr/23398

ggerganov committed 58 days ago
cont : clean-up

ggerganov committed 58 days ago
cont : clean-up

ggerganov committed 58 days ago
cont : fix handling of unused tensors

ggerganov committed 58 days ago
cont : fix undefined

ggerganov committed 58 days ago
fix typo

CISC committed 58 days ago
cont : enable gemma4 graph reuse

ggerganov committed 58 days ago
cont : fix assert

ggerganov committed 58 days ago
cont : fix quantized cache

ggerganov committed 58 days ago
cont : fix names

ggerganov committed 58 days ago
cont : fix names

ggerganov committed 58 days ago
cont : add reference for draft positions

ggerganov committed 58 days ago
cont : fix multi-modality

ggerganov committed 58 days ago
cont : add comment about ctx_src

ggerganov committed 58 days ago
cont : clean-up server fit logic

ggerganov committed 58 days ago
cont : clean-up llama_context

ggerganov committed 58 days ago
py : fix names

ggerganov committed 58 days ago
cont : rename ctx_src -> ctx_other

ggerganov committed 58 days ago

Loading