Kimi Linear chunk size = 16 #19827
ggerganov
merged
26 commits into
ggml-org:master
from
ymcki:dn
models : add llm_build_delta_net_base
fa44de74
cont : keep qwen35 and qwen35moe graphs intact
403e78ea
cont : add comments [no ci]
2371dfb5
add kimi linear to delta-net-base
cff8f601
sync to b8057
b0594c9f
removed unnecessary ggml_cont from g_exp_t
6c765ebf
removed ggml_cont from g_diff_exp_t. moved ggml_cont for o to kimi-li…
a93bcc4d
removed unnecessary diag mask
7b26805f
cont : simplify
4a6393ef
cont : avoid graph splits
c07977ae
scale q after mul instead of beginning
117763ed
Merge branch 'ggml-org:master' into dn
6dad4378
Merge branch 'master' of github.com:ymcki/llama.cpp into dn
df269dc7
scale q after mul instead of beginning
1cea24dc
scale q after mul instead of beginning
a6fa6d50
identical ppl
6432f95b
cont : fix scale and decay mask
de6a8420
minor : remove TODO
23cccea2
block implementation for kda
18d7b2ca
block implementation for kda
09f0baf7
remove space at the end of line 101
ac46b38d
concat+pad
ec25a26c
pad+binary row concat
8c96f826
chunk size 16 for kda
aa5b8169
Merge branch 'ggml-org:master' into dn
099645a7
ggerganov
commented
on 2026-03-02
removed minor differences to master
f90c5850
ggerganov
approved these changes
on 2026-03-02
Login to write a write a comment.
Login via GitHub