ngxson/llama.cpp

Pull Requests Commits

ngxson committed 19 hours ago

a5f27252

server: model-agnostic impl of systemone/decision API

ngxson committed 1 day ago

f30811c8

ngxson committed 1 day ago

aa07784b

ngxson committed 1 day ago

54a0c5da

ngxson committed 1 day ago

0d1cb9b5

support lev & kev

ngxson committed 1 day ago

8536d4fb

ngxson committed 1 day ago

82f7f147

add openjev tiny model for testing

ngxson committed 1 day ago

7c760c1e

add vision support

ngxson committed 1 day ago

afabc13b

Merge branch 'master' into xsn/decision

ngxson committed 1 day ago

9efdb2f5

add docs, imorove UX a bit

ngxson committed 1 day ago

93ed62dc

support shared prompt prefix

ngxson committed 1 day ago

d244cdcf

improve conversion script

ngxson committed 1 day ago

85eae8ae

add server code

ngxson committed 1 day ago

1b89754b

ngxson committed 1 day ago

654a5622

ngxson committed 1 day ago

d2d6995b

ci: fix Fusion / metal by updating the qwen4exp baseline (#29812)

ServeurpersoCom committed 1 day ago

Verified e358d591

llama : clamp kpool re-pool bound to existing pools (#29805)

ggerganov committed 1 day ago

Verified 81e39ad3

init conversion

ngxson committed 1 day ago

55e94c9a

hexagon: shared strided DMA copy for CPY and CONCAT, any-dim CONCAT via DMA (#29685)

njsyw1997 committed 1 day ago

Verified dcd387a4

server: return HTTP 400 for invalid embedding requests (#29060)

SamMalayek committed 1 day ago

Verified d775ebf3

convert : write Gemma embedding scale for DFlash drafts (#29802)

kabu1204 committed 1 day ago

Verified 2b36825c

cuda : route sm70 to the Turing MMVQ nwarps table (#29753)

tkittich committed 1 day ago

Verified 42d95816

metal : release temporary private transfer buffers (#29777)

mvanlamz committed 1 day ago

Verified 13b4d713

webgpu: add bfloat16 support for MUL_MAT/MUL_MAT_ID/GET_ROWS- #29358 (#29358)

yomaytk committed 1 day ago

Verified 4b1622af

llama : fix invalid assert in recurrent memory (#29799)

ggerganov committed 1 day ago

Verified 869034b4

CUDA: Handle compute type for NVFP4 on cublass path (#29173)

ynankani committed 1 day ago

Verified b56f34ab

Qwen4Exp: add MTP (#29761)

am17an committed 1 day ago

Verified c061df19

llama: fix qwen4exp (#29751)

am17an committed 1 day ago

Verified 66e0c17e

CUDA: Make CCCL configurable + pin it to 3.4.3 for CI jobs (#29792)

ORippler committed 1 day ago

Verified 76776785

Older