Gemma4 MTP#17

Open

am17an wants to merge 51 commits into

Commits on May 19, 2026

Commits on May 20, 2026

Commits on May 21, 2026

app : show version (#23426)
angt
authored
hexagon: ssm-conv fix for large prompts (#23307)

tboinovski1
and
max-krasnyansky
authored
llama-graph: fix null-buffer crash in llm_graph_input_attn_kv_iswa for SWA-only models (#23131)
ssfdre38
authored
ggml : Check the right iface method before using the fallback 2d get (#23306)
TheBlueMatt
authored
ui: Improve Git Hooks for UI development (#23403)

allozaur
and
ServeurpersoCom
authored
doc: fix spec mtp typo (#23435)
ruixiang63
authored
vocab : add Carbon-3B (HybridDNATokenizer) support (#23410)

kashif
and
CISC
authored
mtp: use inp_out_ids for skipping logit computation (#23433)
am17an
authored
app : add batched-bench, fit-params, quantize & perplexity (#23459)
angt
authored
server: re-inject subcommand when router spawns children under unified binary (#23442)
ServeurpersoCom
authored
server : free draft/MTP resources on sleep to fix VRAM leak (#23461)
am17an
authored
metal : optimize concat kernel and fix set kernel threads (#23411)
ggerganov
authored
server: expose prompt token counts in /slots endpoint (#23454)
ScrewTSW
authored
tests : move save-load-state from examples to tests (#23336)
ggerganov
authored
fix(flash-attn): replace f32 with kv_type and q_type (#23372)
Constannnnnt
authored
vulkan: fuse snake activation (mul, sin, sqr, mul, add) (#22855)
ServeurpersoCom
authored
Update WebGPU support and add link to blog/demo (#23483)
reeselevine
authored
cmake : remove STATIC from impl libraries, enable LLAMA_BUILD_APP by default (#23462)
ggerganov
authored
CUDA: fix PDL CC check for JIT compilation (#23471)
JohannesGaessler
authored

Commits on May 22, 2026

Commits on May 23, 2026

add Q rot when cache is quantized
am17an
committed