ggerganov/llama.cpp

Pull Requests Commits

fix(ui): take the agentic turns default from DEFAULT_AGENTIC_CONFIG

allozaur committed 14 hours ago

4fa59473

jinja : write nan and inf in tojson as json.dumps does (#30483)

ykhrustalev committed 14 hours ago

Verified 82e1e2fd

vulkan : fix concurrent first device init (#29766)

losewayy committed 15 hours ago

Verified 47f5a64a

sycl : replace oneAPI package by DLE+oneDNN package (#30131)

arthw committed 16 hours ago

Verified e36df604

opencl: fix OOB tail access for adreno noshuffle gemv kernels (#30421)

lhez committed 16 hours ago

Verified 9166ceaf

jinja : fix loop body variable scoping (#30533)

Ninja3047 committed 16 hours ago

Verified 2130ca84

ggml-cpu : optimize pad_f32 (#30526)

angt committed 16 hours ago

Verified 23f83e05

unicode : reject overlong UTF-8 and out-of-range codepoints (#30485)

simoneiacomino committed 16 hours ago

Verified 2ce20d22

llama : split llama_prefetch() into merge and prefetch steps (#30532)

angt committed 16 hours ago

Verified 68ac7657

opencl: apply flash attn softcap before mask (#30546)

lhez committed 16 hours ago

Verified 77c1f399

vulkan : copy strided f16 KV once on AMD so reads spread across memory channels (#27703)

Nathanw1014 committed 16 hours ago

Verified 733f57a9

cpu: add F16 SWIGLU_OAI reference implementation and validation (#29200)

cqderek committed 16 hours ago

Verified 5a638aed

metal : add BF16 variant of the tensor API flash attention kernel (#30491)

ethanhq committed 16 hours ago

Verified 6e66cb5d

kv-cache: honor PARTIAL_ONLY in the plain KV cache (#30514)

ServeurpersoCom committed 19 hours ago

Verified e467e5b5

graph: keep the graph non empty without embeddings (#30522)

ServeurpersoCom committed 19 hours ago

Verified 1ddc2625

server: save only restorable slot states to the prompt cache (#30498)

ServeurpersoCom committed 19 hours ago

Verified 2639a3b4

spec : add lilicorr support (#30506)

frozenblade1224 committed 19 hours ago

Verified 843e7dcc

vulkan: Use L2-friendly workgroup scheduling for cm2 mulmat (#30215)

0cc4m committed 20 hours ago

Verified 3e390af1

opencl: support CLAMP on non-contiguous views (#30186)

wanghqc committed 1 day ago

Verified a0e6dc73

devops: incorrect COPY instruction (#30529)

stormyhs committed 1 day ago

Verified 23deec02

gguf-py : detect nfc normalization (#30540)

CISC committed 1 day ago

Verified 683bbc0d

sycl: IQ3 code reorder (#29107)

anantshri committed 1 day ago

Verified 529ac893

ggml-cpu : keep the FP16->FP32 lookup table on x86 with F16C (#30342) (#30343)

Tough-Respawn committed 1 day ago

Verified 685f7089

docs: update IBM zDNN backend docs (#30527)

taronaeo committed 1 day ago

Verified c114bd4c

ggml-cpu : size set_rows conversion buffer for all threads (#30454)

am17an committed 1 day ago

Verified 5cdadfc3

cuda, hip: GDN chunked kernel (#29353)

am17an committed 1 day ago

Verified 62df1815

skill: Update `code-review` skill for UI development (#30509)

allozaur committed 1 day ago

Verified 37d6d653

ggml: fix conv_1d result layout for batched input (#28738)

ServeurpersoCom committed 1 day ago

Verified 6868a52c

common : add --spec-draft-n-ubatch (#30459)

ggerganov committed 1 day ago

Verified 3fb47677

docs : clarify installation recommendation for Intel oneAPI Toolkit (#30407)

alosslessdev committed 1 day ago

Verified 318a471a

Older