ggerganov/llama.cpp

Pull Requests Commits

metal : reuse graphs

ggerganov committed 120 days ago

Verified bf8b3901

llama-bench : add graph reuse parameter

ggerganov committed 120 days ago

Verified 0d2038f9

llama : reuse compute graphs

ggerganov committed 122 days ago

Verified 76681e3c

eval-callback : check for empty input (#14539)

ggerganov committed 122 days ago

Verified bac8bed2

test-backend-ops: add support for specifying output format (#14368)

yeahdongcn committed 122 days ago

Verified b81510a7

metal : disable fast math in all quantize kernels (#14528)

ggerganov committed 123 days ago

Verified ef797db3

batch : add optional for sequential equal split (#14511)

ggerganov committed 123 days ago

Verified 67d1ef23

graph : prepare for 4D mask (#14515)

ggerganov committed 123 days ago

Verified 7b50f7c0

batch : add n_used count (#14512)

ggerganov committed 123 days ago

Verified c79184d2

CANN: Replace aclrtMemsetSync with aclnnInplaceZero operator (#14002)

luyhcsu committed 123 days ago

Verified 499a8f5a

ggml : implement GEGLU_ERF and GEGLU_QUICK ops (#14445)

CISC committed 123 days ago

Verified 28657a82

opencl : broadcast for soft_max (#14510)

lhez committed 124 days ago

Verified bee28421

vulkan: support mixed/deepseekR1 FA head sizes (#14509)

jeffbolznv committed 124 days ago

Verified 2b72bede

ggml: backward pass for split swiglu (#14483)

JohannesGaessler committed 124 days ago

Verified c8c4495b

Fix conditional enabling following arch checks for ggml-sycl (#14504)

s-Nick committed 124 days ago

Verified 7b63a71a

convert : correct gemma 3n conversion (#14450)

ngxson committed 124 days ago

Verified 0c2ee38a

kv-cache : use ggml_set_rows (#14285)

ggerganov committed 124 days ago

Verified a70c8a0c

ggml : fix FA mask dim 2 and 3 (#14505)

ggerganov committed 124 days ago

Verified 9067487c

ggml : remove kompute backend (#14501)

ggerganov committed 124 days ago

Verified d4cdd9c1

CUDA: add dynamic shared mem to softmax, refactor general usage (#14497)

am17an committed 124 days ago

Verified 55c2646b

gguf-py : add support for chat template jinja files (#14508)

CISC committed 125 days ago

Verified e75ba4c0

llama : initial Mamba-2 support (#9126)

compilade committed 125 days ago

Verified 5d46babd

ggerganov committed 125 days ago

e17991c4

ggml : add version function to get lib version (ggml/1286)

danbev committed 125 days ago

c46944aa

Set RPATH to "@loader_path" / "$ORIGIN" to ensure executables and dynamic libraries search for dependencies in their origin directory. (#14309)

rotemdan committed 125 days ago

Verified f3ed38d7

CUDA: add softmax broadcast (#14475)

am17an committed 125 days ago

55a1c5a5

CUDA: broadcasting for FlashAttention mask (#14500)

JohannesGaessler committed 125 days ago

12a81af4

vulkan: support softmax/FA batch and broadcast (#14449)

jeffbolznv committed 125 days ago

8875523e

ggml : support bcast ggml_soft_max_ext, ggml_flash_attn_ext (#14435)

ggerganov committed 125 days ago

ec68e84c

opencl : fix possible buffer overflow in dump_tensor (#14490)

jeffzhou2000 committed 125 days ago

Verified 307e79d3

Older