ggerganov/llama.cpp

Pull Requests Commits

cpu: adjust splitting thresholds and add a #define for them

max-krasnyansky committed 1 day ago

2df55ed7

cpu: split large single-row tensors to improve CPU utilization in unary and binary ops during t.g

max-krasnyansky committed 1 day ago

ba97234c

hexagon: MUL_MAT_ID row skipping optimizations and improvements (#31169)

max-krasnyansky committed 1 day ago

Verified caa8191d

llama : move no_alloc out of llama_hparams (#31141)

ggerganov committed 1 day ago

Verified a4699400

models : replace ggml_tensor** out-params with output structs (#31142)

ggerganov committed 1 day ago

Verified 4b851fbe

ggml-cpu : rewrite concat as one row kernel for all types (#31179)

ServeurpersoCom committed 1 day ago

Verified 4d71b42f

vulkan: rms norm optimization using subgroup reductions (#29882)

virajwad committed 1 day ago

Verified 70c5f263

sycl: template the plain MMVQ launch triples (#31013)

Titaniumtown committed 1 day ago

Verified 5cd62ffe

sycl: template the per-quant concat impls (#31008)

Titaniumtown committed 1 day ago

Verified 38645c2a

sycl: add set_tensor_2d for faster tensor-split loading (#30358)

Captain-Tripps committed 1 day ago

Verified 02269b4c

sycl : bound reorder temp memory to 32 MiB (#31062)

KyleHagy committed 1 day ago

Verified 5320c87f

common: pin model resolution on real HF repo layouts (#30548)

ServeurpersoCom committed 1 day ago

Verified 74b104c9

mtmd: add mtmd_helper_get_capabilities (#31183)

ngxson committed 1 day ago

Verified c6fb3b5b

ggml-cpu : fix unused variables in arm arch (#31180)

CISC committed 1 day ago

Verified 8b28c239

tests : refactor test-llama-archs to not link internal libllama symbols (#30880)

ggerganov committed 1 day ago

Verified fd8f8451

ci : fix race in webgpu memset (#31162)

eapache committed 1 day ago

Verified 9125a2eb

metal : stop overflow from happening in large matrix output offsets (#30072)

feni6 committed 1 day ago

Verified 9b8446b9

ui: UX improvements (#31030)

allozaur committed 1 day ago

Verified dc9e3bed

ui: Sidebar resizing improvements (#31119)

allozaur committed 1 day ago

Verified 69b3cc40

ci : bump server timeout during startup (#31174)

ggerganov committed 2 days ago

Verified 224d94c0

ci : drop the `macos-latest-x64` job (#31175)

ggerganov committed 2 days ago

Verified 48c36110

deepseekv4.1: follow-up review fixes (#31137)

am17an committed 2 days ago

Verified a0742153

cuda : use std::string instead of VLA for archName (#30463)

angt committed 2 days ago

Verified d854d98b

common, grammar: add recursive depth limit, make it a requirement (#31153)

ngxson committed 2 days ago

Verified 48c9f08b

tests : adjust default threads for test-llama-context (#31157)

ggerganov committed 2 days ago

Verified 6050b467

qwen: merge tensors on convert (#31028)

am17an committed 2 days ago

Verified 2280f0fd

ci : disable test-llama-context with fa off in the webgpu (#30977)

yomaytk committed 2 days ago

Verified 1a0b56ce

llama: fix exhausting CPU and MEM caused by dry sequence breaker (#31035)

trunk-trick committed 2 days ago

Verified 63412d87

llama-quant : fix 4D tensor quantization using ggml_nrows() and name-based expert detection (#30820)

anthony-wss committed 2 days ago

Verified 5c965fd8

grammar: limit recursion depth to prevent stack overflow (#29495) (#29755)

Hrushikesh-ramilla committed 2 days ago

Verified f1168ac4

Older