ggerganov/llama.cpp

Pull Requests Commits

llama : ragged ubatches for GDN hybrid models (RFC)

am17an committed 29 minutes ago

5d2d11be

llama : drop the fused GDN probes brought back by the rebase

am17an committed 3 days ago

33a1dbf7

gdn: fix rebase conflicts (missing cparams, etc)

max-krasnyansky committed 3 days ago

207d2c83

hex-gdn: add support for replay

max-krasnyansky committed 3 days ago

ab2fb0da

vulkan: clamp the gated delta net replay count to n_pend

ServeurpersoCom committed 3 days ago

36751006

delta_net: share the replay conv window views across layers

ServeurpersoCom committed 3 days ago

e0987318

webgpu: replay the pending gated delta net tokens in the shader

ServeurpersoCom committed 3 days ago

30062d86

vulkan: replay the pending gated delta net tokens in the shader

ServeurpersoCom committed 3 days ago

b85e15ad

metal: replay the pending gated delta net tokens in the kernel

ServeurpersoCom committed 3 days ago

e6fbf3be

delta_net: drop the graph replay fallback

ServeurpersoCom committed 3 days ago

5b8d8a7b

memory : move rs_max with the sequence when find_slot swaps cells

am17an committed 3 days ago

e93f8d08

GDN: use replay instead of rollback [WIP]

am17an committed 3 days ago

4df118ea

sycl: make oneDNN FA cache thread-safe and memoize runtime failures (#31138)

Titaniumtown committed 3 days ago

Verified 09338b85

sycl : fix AOT for more than 4GB (#31873)

arthw committed 3 days ago

Verified dcbff216

sycl : use cached device-id in set_main_device (#32013)

zhaoyshine committed 3 days ago

Verified f3d7725a

CUDA: add cuda debug buffer pool (#31919)

JoeyPatricio committed 3 days ago

Verified 7021f7e0

opencl: add bin kernel `kernel_gemm_noshuffle_q4_0_f32_32b_trans_ila_a7_bin` (#30947)

shaofeiqi committed 3 days ago

Verified e0949d09

CUDA: skip empty MoE tiles in MMQ stream-k fixup (#32067)

lifeiteng committed 3 days ago

Verified 5d4c8eeb

server: use the moved test model in the router CORS test (#32117)

allozaur committed 3 days ago

Verified dd6b945b

opencl: add decomposed FA for Adreno (#32046)

shaofeiqi committed 3 days ago

Verified e1f7593c

ui: move js sandbox setting out of "developer" (#32054)

ngxson committed 3 days ago

Verified 7bcbc04f

docs: update image link in README.md (#32038)

inimaz committed 3 days ago

Verified dd00306b

ui: keep downloads in Discover during search and fix their progress with a draft (#31467)

allozaur committed 3 days ago

Verified 1d819927

ui: take the logo and favicons from the llama-brand package (#32039)

allozaur committed 3 days ago

Verified bd717e08

server: fix --cors-origins handling + protect against DNS rebinding (#32098)

ngxson committed 3 days ago

Verified 2cbe0ec4

ci: move test models to ggml-org-dev (#32096)

ngxson committed 3 days ago

Verified 664f031f

metal : prevent out-of-bounds reads in SOLVE_TRI kernel (#32103)

realMoana committed 3 days ago

Verified 4180722e

vulkan: support batched tensor copy API (#31418)

0cc4m committed 3 days ago

Verified 055c3c26

vendor : update LibreSSL to 4.4.1 [no ci] (#32106)

cabelo committed 3 days ago

Verified 2656b71a

llama : reuse graph result instances across scheduler reserves (#32089)

ggerganov committed 3 days ago

Verified cb44fdb1

Older