ngxson/llama.cpp

Pull Requests Commits

ngxson committed 25 days ago

d5a08569

bring back ggml_concat as it may not work on other backend

ngxson committed 26 days ago

329112c5

avoid ggml_concat inside loop

ngxson committed 26 days ago

e41d9100

rm unused tensor

ngxson committed 26 days ago

9299ced6

vectorize key_gdiff calculation

ngxson committed 26 days ago

5ec140e0

use std::pair instead of ggml_concat

ngxson committed 26 days ago

f8ad742a

add comments about no cont

ngxson committed 26 days ago

f38fc605

improve g_cs_chunk

ngxson committed 26 days ago

939767c3

rm redundant cont

ngxson committed 26 days ago

e1f8ad25

move mul_mat(k,q) to outside of chunking

ngxson committed 26 days ago

2a39955a

rm 2 redundant transposes

ngxson committed 26 days ago

d96eb69e

Merge branch 'master' into xsn/qwen3next_improve

ngxson committed 26 days ago

033fd273

opencl: add EXPM1 op (#18704)

shaofeiqi committed 26 days ago

Verified 593da7fa

Updates to webgpu get_memory (#18707)

reeselevine committed 27 days ago

Verified 9e41884d

Webui/file upload (#18694)

ServeurpersoCom committed 27 days ago

Verified ec8fd787

cmake: only build cli when server is enabled (#18670)

AsbjornOlling committed 27 days ago

Verified a180ba78

server : fix timing of prompt/generation (#18713)

ggerganov committed 27 days ago

Verified 53eb9435

scripts : pr2wt.sh reset to remote head (#18695)

ggerganov committed 27 days ago

Verified d3435efc

server : use different seeds for child completions (#18700)

ggerganov committed 27 days ago

Verified f5f8812f

common: support remote preset (#18520)

ngxson committed 27 days ago

Verified 8ece3836

llama: use host memory if device reports 0 memory (#18587)

taronaeo committed 27 days ago

Verified 046d5fd4

ggml-webgpu: Fix GGML_MEM_ALIGN to 8 for emscripten. (#18628)

yomaytk committed 28 days ago

Verified 480160d4

ggml webgpu: initial flashattention implementation (#18610)

reeselevine committed 28 days ago

Verified 15bff84b

fix missing reshape

ngxson committed 28 days ago

c77001f2

vulkan: fix push constant size for quantize_q8_1 (#18687)

jeffbolznv committed 28 days ago

Verified 2524c261

vulkan: optimize ssm_scan (#18630)

jeffbolznv committed 28 days ago

Verified cb14b069

vendor : update cpp-httplib to 0.30.0 (#18660)

angt committed 28 days ago

Verified 55abc393

scripts : support chaining commands in pr2wt.sh (#18671)

ggerganov committed 28 days ago

Verified f2f6c880

metal : add MoE kernel specialization for ne20=5 (#18667)

dororodoroddo committed 28 days ago

Verified 945bf106

llama-fit-params: free memory target per device (#18679)

JohannesGaessler committed 28 days ago

Verified 64848deb

Older