ggml-org/llama.cpp

Pull Requests Commits

address feedback

0cc4m committed 45 days ago

f3ba33ec

cuda: read memory through NVML if available to avoid initializing a context

0cc4m committed 46 days ago

3c4d2b75

convert : minor fixes for numpy 2.x (#23571)

CISC committed 46 days ago

Verified 5d246a79

cmake : fix ui build (#23592)

aldehir committed 46 days ago

Verified 63248fc3

server: add margin for draft model for `fit` (#23485)

am17an committed 46 days ago

Verified 83eebe9d

TP: fix entirely zero-sized slices per device (#23525)

JohannesGaessler committed 46 days ago

Verified fff63b51

opencl: batch profiling to improve speed and prevent memory leaks (#23495)

shaofeiqi committed 46 days ago

Verified f3061116

hexagon: apply repl optimization in flash attn softmax as #22993 (#23455)

njsyw1997 committed 47 days ago

Verified 1c0f6db5

snapdragon: update windows toolchain to use hsdk v6.6.0.0 (#23552)

aparmp-quic committed 47 days ago

Verified cec51c7a

cmake/ui : refactor the build (#23352)

aldehir committed 47 days ago

Verified b22ff4b7

requirements : bump torch to 2.11.0 (#23503)

adityasingh2400 committed 47 days ago

Verified c0c7e147

model : add NVFP4 MTP scale tensors (#23563)

michaelw9999 committed 47 days ago

Verified b0df4c0c

ggml : Check the right iface method before using the fallback 2d get (#23514)

dskwe committed 47 days ago

Verified a4974763

vulkan: fix windows find_package of SPIRV-Headers (#23215)

jeffbolznv committed 47 days ago

Verified 95405ac6

opencl: generalize Adreno MoE kernels on M (#23449)

shawngu-quic committed 48 days ago

Verified 0f3cb3fc

server: only parse empty msg if continuing an assistant msg (#23506)

aldehir committed 48 days ago

Verified 1acee6bf

perplexity : fix integer overflow (#23496)

fairydreaming committed 48 days ago

Verified ef570f63

SYCL: improve MoE prefill throughput (#23142)

sanmai committed 48 days ago

Verified cc9e3312

sycl : Level Zero detection in ggml_sycl_init (#23097)

sanmai committed 48 days ago

Verified bcfd1989

SYCL : gated_delta_net K>1 (#23174)

karavayev committed 48 days ago

Verified 56f16f23

SYCL: add BF16 to DMMV kernel path (~4x tg speedup on Intel Arc) (#21580)

PMZFX committed 48 days ago

Verified 8cc67efc

docs: Update documentation with Granite 4.0/4.1 (#23404)

jesus-talavera-ibm committed 48 days ago

Verified 95feeab5

ggml-zendnn : add Q8_0 quantization support (#23414)

z-sachin committed 48 days ago

Verified 99d4026b

cmake : build router app only during standalone builds (#23521)

fairydreaming committed 48 days ago

Verified 9c92e96a

vocab : fix HybridDNA tokenizer (#23466)

kashif committed 48 days ago

Verified afcda09d

cmake : add install() for impl libraries + fix apple builds (#23511)

ggerganov committed 48 days ago

Verified bbce619a

CUDA: fix PDL CC check for JIT compilation (#23471)

JohannesGaessler committed 49 days ago

Verified 4f0e43da

cmake : remove STATIC from impl libraries, enable LLAMA_BUILD_APP by default (#23462)

ggerganov committed 49 days ago

Verified bb28c1fe

Update WebGPU support and add link to blog/demo (#23483)

reeselevine committed 49 days ago

Verified ee7c3057

vulkan: fuse snake activation (mul, sin, sqr, mul, add) (#22855)

ServeurpersoCom committed 49 days ago

Verified 47c0eda9

Older