ggerganov/llama.cpp

Pull Requests Commits

fix sparse FA fp16 underflow bug

0cc4m committed 4 days ago

9731be49

reduce occupancy on RDNA4 for large attention head sizes

0cc4m committed 4 days ago

d52056be

0cc4m committed 4 days ago

7f14294d

fix skip condition

0cc4m committed 4 days ago

295921ab

only store pv in sfsh if necessary

0cc4m committed 4 days ago

f6e5934e

fix nvidia hang, tune AMD

0cc4m committed 4 days ago

d56b44d6

replace kvsep with direct load mode

0cc4m committed 4 days ago

320e68a0

move flip shader into own file

0cc4m committed 4 days ago

61a57498

0cc4m committed 4 days ago

d5219c8f

vulkan: add flipped FA path, each subgroup owns 16 Q rows

0cc4m committed 4 days ago

b91f7de7

limit q_in_regs to prefill

0cc4m committed 4 days ago

cdf118a7

clean up and fixes

0cc4m committed 4 days ago

aee457d0

reduce shmem use through buffer reuse

0cc4m committed 4 days ago

c02faf8f

vulkan: optimize Flash Attention cm1 for RDNA3 and RDNA4

0cc4m committed 4 days ago

075deb79

ui: build the reasoning menu from the template effort levels (#31273)

ServeurpersoCom committed 4 days ago

Verified e486a4b5

tests : adjust constants to reduce irrelevant warnings (#31254)

ggerganov committed 4 days ago

Verified f1a7dea9

cuda : add top-k MoE fusion for 384 experts (#31288)

developer0hye committed 4 days ago

Verified a7af149a

vulkan: clamp A prefetch to end_k in coopmat1 int8 matmul (#30283)

SelfRef committed 4 days ago

Verified 2864937a

chat : stop qwen3-coder reasoning and content before template separators (#30557)

7konigsberg committed 4 days ago

Verified a71c00f0

ui : show save and reset settings buttons only when relevant (#31248)

allozaur committed 5 days ago

Verified f4d8af60

ui: Support OGG audio attachments by converting them to WAV (#31237)

allozaur committed 5 days ago

Verified 02d502f3

ui: Fix draft for new message disappearing after using "Add System Message" action (#31214)

allozaur committed 5 days ago

Verified b1756b23

ui : Search dialog (#31173)

allozaur committed 5 days ago

Verified 6886628c

server : add realtime conversation and transcription on /v1/realtime (#31269)

tdakhran committed 5 days ago

Verified 8ca4060b

model : add LiquidAI/d1-2B and LiquidAI/d1-omni-2B (#31275)

tdakhran committed 5 days ago

Verified d05706b4

server: expose the reasoning effort levels of the chat template (#31271)

ServeurpersoCom committed 5 days ago

Verified 1b4554ef

tests: fix CI riscv timeout in test-llama-context (#31265)

ServeurpersoCom committed 5 days ago

Verified a9191799

HIP/CUDA: remove pragma unroll from loops in conv2d where itteration count cant be infered at compile time (#31255)

IMbackK committed 5 days ago

Verified 338fce65

model : add LiquidAI/LFM3-Audio-2B and LiquidAI/LFM3-Omni-2B (#31257)

tdakhran committed 5 days ago

Verified 19ff817f

common : support deferred tools and tool_reference content parts (#31256)

tdakhran committed 5 days ago

Verified eb7818c0

Older