ngxson/llama.cpp

Pull Requests Commits

llama : correct rms norm for llama 4

ngxson committed 1 year ago

7c6ae54a

ggerganov committed 1 year ago

47ba87d0

tests : fix init order (#0)

ggerganov committed 1 year ago

1d2b6134

ggerganov committed 1 year ago

eb420e11

ggml: don't include arm_neon.h when using CUDA 12 with ARM Neon (ggml/1187)

cmdr2 committed 1 year ago

cb79c2e7

ggml : add bilinear upscale support (ggml/1185)

slaren committed 1 year ago

fe92821e

ggml : add more generic custom op, remove deprecated custom ops (ggml/1183)

slaren committed 1 year ago

459895c3

scripts : fix sync-ggml-am.sh

ggerganov committed 1 year ago

e4bf72d6

llava : introduce libmtmd (#12849)

ngxson committed 1 year ago

Verified 8b9cc7cd

convert : ability to lazy-load safetensors remotely without downloading to disk (#12820)

ngxson committed 1 year ago

Verified 64eda5de

CANN: Support more ops (#12841)

noemotiovon committed 1 year ago

Verified fe5b78c8

Fixes #12823 (#12830)

mehendarkarprajwal committed 1 year ago

Verified 11d07e1e

docker : added all CPU to GPU images (#12749)

rudiservo committed 1 year ago

Verified b0091ecc

ggml-cpu-impl.h: do not redefine bool on POWER9 (#12856)

pkubaj committed 1 year ago

Verified 31f7803b

ggml-impl.h: fix build on POWER9 (#12855)

pkubaj committed 1 year ago

Verified 2391506a

llama : Support Qwen3 and Qwen3MoE (#12828)

bozheng-hit committed 1 year ago

Verified d3bd7193

musa: enable freediskspace for docker image build (#12839)

yeahdongcn committed 1 year ago

Verified d9a63b2f

sycl: update documentation to use -no-cnv (#12845)

Rbiessy committed 1 year ago

Verified 8ed71242

ci: detach common from the library (#12827)

pminev committed 1 year ago

Verified 381603a7

clip : do not print ftype (#12832)

ngxson committed 1 year ago

Verified 65a69e6e

readme : add rpc backend (#12842)

ggerganov committed 1 year ago

Verified 47277d6d

CANN: Support Opt CONV_TRANSPOSE_1D and ELU (#12786)

noemotiovon committed 1 year ago

Verified 6e1c4ceb

vulkan: In coopmat2 mmq, load q4_k/q5_k scales through shared memory (#12833)

jeffbolznv committed 1 year ago

Verified 0090950f

vulkan: Use fp16 for the flash attention P*V multiplication (#12783)

jeffbolznv committed 1 year ago

Verified 7ecd780b

cuda : add f32 to bf16 copy op (#12806)

CISC committed 1 year ago

Verified 7538246e

llava: improve clip_ctx destructor to not memleak load_image_size (#12834)

mattjcly committed 1 year ago

Verified b32efad2

llama : fix FA when KV cache is not used (i.e. embeddings) (#12825)

ggerganov committed 1 year ago

Verified a19b5cef

server : fix thread.join() on exit (#12831)

ngxson committed 1 year ago

Verified 78a1ba0a

llava: add more helper functions to check projector types in clip context (#12824)

dm4 committed 1 year ago

Verified 2dabf759

arg : Including limits file on AIX (#12822)

mehendarkarprajwal committed 1 year ago

Verified 1d343b40

Older