ngxson/llama.cpp

Pull Requests Commits

ngxson committed 22 days ago

a5f27252

server: model-agnostic impl of systemone/decision API

ngxson committed 22 days ago

f30811c8

ngxson committed 22 days ago

aa07784b

ngxson committed 23 days ago

54a0c5da

ngxson committed 23 days ago

0d1cb9b5

support lev & kev

ngxson committed 23 days ago

8536d4fb

ngxson committed 23 days ago

82f7f147

add openjev tiny model for testing

ngxson committed 23 days ago

7c760c1e

add vision support

ngxson committed 23 days ago

afabc13b

Merge branch 'master' into xsn/decision

ngxson committed 23 days ago

9efdb2f5

add docs, imorove UX a bit

ngxson committed 23 days ago

93ed62dc

support shared prompt prefix

ngxson committed 23 days ago

d244cdcf

improve conversion script

ngxson committed 23 days ago

85eae8ae

add server code

ngxson committed 23 days ago

1b89754b

ngxson committed 23 days ago

654a5622

ngxson committed 23 days ago

d2d6995b

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

ServeurpersoCom committed 23 days ago

Verified e358d591

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

ggerganov committed 23 days ago

Verified 81e39ad3

init conversion

ngxson committed 23 days ago

55e94c9a

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

njsyw1997 committed 23 days ago

Verified dcd387a4

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

SamMalayek committed 23 days ago

Verified d775ebf3

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

kabu1204 committed 23 days ago

Verified 2b36825c

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

tkittich committed 23 days ago

Verified 42d95816

metal : release temporary private transfer buffers (#29777)

mvanlamz committed 23 days ago

Verified 13b4d713

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

yomaytk committed 23 days ago

Verified 4b1622af

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

ggerganov committed 23 days ago

Verified 869034b4

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

ynankani committed 23 days ago

Verified b56f34ab

Qwen4Exp: add MTP (#29761)

am17an committed 23 days ago

Verified c061df19

llama: fix qwen4exp (#29751)

am17an committed 23 days ago

Verified 66e0c17e

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

ORippler committed 23 days ago

Verified 76776785

Older