model : add GLM-5.3-Flash (glm5next) - #27752

Open

eauchs wants to merge 11 commits into

ggml-org:masterfrom

eauchs:glm5next/add-glm-5.3-flash

Commits on Sep 3, 2026

model : add GLM-5.3-Flash (glm5next)
eauchs
committed
6174c91
convert : keep text_config merge from clobbering architectures
eauchs
committed
e972e2c
llama : accept indexer.kpool as a fallback for indexer.block_size
eauchs
committed
dca65ec
llama : select whole k-pools in the glm5next DSA indexer
eauchs
committed
8ee6fcf
llama : add the glm5next MTP (NextN) draft graph
eauchs
committed
38d16c7
llama : extend llama_memory_hybrid_idx for the glm5next DSA path
eauchs
committed
347f2e3
llama : chunk the glm5next DSA indexer over tokens

matteoscalabrini
authored and
eauchs
committed
dba3221
llama : skip foreign-sequence cells when building the glm5next k-pools
eauchs
committed
0ccfa15
llama : glm5next fold (d, n_pool) into one ne[1] ahead of the pool softmax
eauchs
committed
c6cbbf9
llama : glm5next declare ssm_a as NOSCAN
eauchs
committed
dd89075
llama : glm5next adapt to master's n_kv_max attn API and per-layer n_ff_exp
eauchs
committed
1d0c76f