ochafik/llama.cpp

Pull Requests Commits

Set GLM4 blk.*.attn_output.weight, kqv_out-* matmul to GGML_PREC_F32 to fix infinity values in output (#13639)

0cc4m committed 1 year ago

Verified c9c64dee

metal : fix typo in FA kernel comments (#13651)

ggerganov committed 1 year ago

Verified c00a2634

kv-cache : add SWA support (#13194)

ggerganov committed 1 year ago

Verified e298d2fb

CANN: Update CANN model support (#13162)

bachelor-dou committed 1 year ago

Verified f0adb80b

sycl : Overcoming workaround for mmap() allocation on Windows (#13482)

s-Nick committed 1 year ago

Verified f7c9429c

common : add load_progress_callback (#13617)

psocolovsky committed 1 year ago

Verified 1dfbf2cf

Vulkan: Add f32 accumulator support to quantized mul mat to fix GLM4 32B incoherence (#13607)

0cc4m committed 1 year ago

Verified 8960efd0

sycl : backend documentation review (#13544)

Alberto Cabrera Pérez committed 1 year ago

Verified 725f23f1

mtmd : add vision support for llama 4 (#13282)

ngxson committed 1 year ago

Verified 92ecdcc0

ci : upgraded oneAPI version in SYCL workflows and dockerfile (#13532)

Alberto Cabrera Pérez committed 1 year ago

Verified f71f40a2

ggerganov committed 1 year ago

d30cb5a7

mnist: fix segmentation fault (ggml/1227)

JohannesGaessler committed 1 year ago

6c35981a

ggml : fix apple OS check in ggml_print_backtrace (ggml/1229)

slaren committed 1 year ago

8b5e19ae

ggml : Fix missing backtrace on Linux (ggml/1228)

danielzgtg committed 1 year ago

60aea028

fix: check model pointer validity before use (#13631)

D2hugging committed 1 year ago

Verified 9c55e5c5

CANN: Support MOE Model MUL_MAT_ID (#13042)

noemotiovon committed 1 year ago

Verified 33d7aed4

server : added --no-prefill-assistant flag (#13608)

isaac-mcfadyen committed 1 year ago

Verified 6a2bc8bf

cmake: use the current build config for vulkan-shaders-gen (#13595)

giladgd committed 1 year ago

Verified e3a7cf6c

parallel : add option for non-shared and larger prompts (#13598)

ggerganov committed 1 year ago

Verified 518329b2

vulkan: move common FA code to flash_attn_base.comp (#13556)

jeffbolznv committed 1 year ago

Verified 2f5a4e1e

vulkan: use scalar FA rather than coopmat2 when N==1 (#13554)

jeffbolznv committed 1 year ago

Verified 4f41ee11

llguidance : official v0.7.20 release (no actual changes) [noci] (#13594)

CoffeeVampir3 committed 1 year ago

Verified 3e0be1ca

server : do not return error out of context (with ctx shift disabled) (#13577)

ngxson committed 1 year ago

Verified 6aa892ec

webui : improve accessibility for visually impaired people (#13551)

ngxson committed 1 year ago

Verified aea9f8b4

readme : add list of dependencies and their license (#13591)

ngxson committed 1 year ago

Verified 06c1e4ab

releases : use arm version of curl for arm releases (#13592)

slaren committed 1 year ago

Verified 415e40a3

metal : add FA-vec kernel for head size 64 (#13583)

ggerganov committed 1 year ago

Verified 654a6779

llama : print hint when loading a model when no backends are loaded (#13589)

slaren committed 1 year ago

Verified 5364ae4b

ci : add ppc64el to build-linux-cross (#13575)

CISC committed 1 year ago

Verified 7c07ac24

sycl : fixed compilation warnings (#13582)

lslusarczyk committed 1 year ago

Verified 0a338ed0

Newer Older