ggml-org/llama.cpp

Pull Requests Commits

Apply suggestion from @ggerganov (src->buffer to buf_src) v2

aendk committed 25 days ago

Verified b9b56b01

Apply suggestion from @ggerganov (src->buffer to buf_src)

aendk committed 25 days ago

Verified 05c74eae

Simplifies synchronizations to adhere to `saaasg` pattern.

aendk committed 25 days ago

Verified 84252009

Corrects initialization of ggml_backend_sync_mode in

aendk committed 25 days ago

Verified 2789c1b3

Reintroduces stricter check for CPU->CUDA backend async copy via

aendk committed 25 days ago

Verified e03fb8ee

Makes opt-in to relax use of explicit syncs more general. Backends like

aendk committed 25 days ago

Verified bba41184

aendk committed 25 days ago

Verified 362934a9

Relax requirement of checks in async CUDA copies from backend and buffer type to just buffer type, to avoid linking issues

aendk committed 25 days ago

Verified 5a77ac71

Reworked backend detection in ggml-backend.cpp to avoid linking

aendk committed 25 days ago

Verified 5fba5961

Adds macro guards to allow compilation in non-CUDA builds

aendk committed 25 days ago

Verified 0ae8664b

Exchanges synchronous copy with async copy function.

aendk committed 25 days ago

Verified 1f959c5c

Adds function to relax sync requirements between input copies on

aendk committed 25 days ago

Verified a187cbdb

Adds CPU-to-CUDA copy capability to

aendk committed 25 days ago

Verified cb39afd2

convert : move experts permutation from Qwen2MoeModel to Qwen3VLMoeTextModel (#19445)

pwilkin committed 25 days ago

Verified 854b09f0

tts : fix typos in README.md [no ci] (#19463)

danbev committed 25 days ago

Verified 66d403c4

CANN: Remove unnecessary wrapper for `gml_backend_buft_is_cann` (#18968)

rauletorresc committed 25 days ago

Verified f0bfe54f

CANN: implement quantized MUL_MAT_ID for MoE models (#19228)

hipudding committed 25 days ago

Verified 52e38faf

cuda : extend GGML_OP_PAD to work with non-cont src0 (#19429)

ggerganov committed 25 days ago

Verified a0d58553

chat: fix case where template accepts type content only (#19419)

ngxson committed 25 days ago

Verified 98e57ca4

mtmd: Implement tiling for LFM2-VL (#19454)

tdakhran committed 25 days ago

Verified 262364e3

Server: log when converting requests to chat completions format (#19457)

openingnow committed 25 days ago

Verified 820ebfa6

spec : remove check rate (#19377)

srogmann committed 25 days ago

Verified 292f6908

ci : add metal server workflows (#19293)

ggerganov committed 25 days ago

Verified 81ddc60c

revert : "[Model] Qwen3.5 dense and MoE support (no vision) (#19435)" (#19453)

ggerganov committed 25 days ago

Verified 972f323e

ggml-virtgpu: add backend documentation (#19354)

kpouget committed 25 days ago

Verified f5e7734f

cmake : add variable to skip installing tests (#19370)

WhyNotHugo committed 26 days ago

Verified 1e8924fd

[Model] Qwen3.5 dense and MoE support (no vision) (#19435)

pwilkin committed 26 days ago

Verified 39bf692a

CUDA: Fix non-contig rope (#19338)

ORippler committed 26 days ago

Verified e06088da

rpc : update from common.cpp (#19400)

angt committed 27 days ago

Verified 5fa1c190

server : improve context checkpoint logic (#19408)

ggerganov committed 27 days ago

Verified eb449cdf

Older