ggerganov/llama.cpp

Pull Requests Commits

llama : store non-RoPEd K cache (WIP)

ggerganov committed 1 year ago

Verified 784d14ed

metal : add rope_f16 kernel + optimize cpy kernels

ggerganov committed 1 year ago

Verified fad56936

ggml : ggml_rope now takes a vector with positions instead of n_past

ggerganov committed 1 year ago

Verified 1fb033fd

llama : replace ggml_diag_mask_inf with ggml_add (custom -inf mask)

ggerganov committed 1 year ago

Verified 3b4bab6a

tests : verify that RoPE is "additive"

ggerganov committed 1 year ago

Verified c5df72e8

common : do not use GNU zero-length __VA_ARGS__ extension (#3195)

cebtenzzre committed 1 year ago

Verified 80291a1d

metal : fix bug in soft_max kernels (out-of-bounds access) (#3194)

ggerganov committed 1 year ago

Verified c6f1491d

convert : make ftype optional in simple scripts (#3185)

cebtenzzre committed 1 year ago

Verified e3d87a6c

sync : ggml (Metal F32 support + reduce ggml-alloc size) (#3192)

ggerganov committed 1 year ago

Verified 8c00b7a6

cmake : fix building shared libs for clang (rocm) on windows (#3176)

Engininja2 committed 1 year ago

Verified 7e50d34b

flake : use pkg-config instead of pkgconfig (#3188)

kurnevsky committed 1 year ago

Verified 235f7c19

metal : relax conditions on fast matrix multiplication kernel (#3168)

ggerganov committed 1 year ago

Verified a51b6876

cmake : fix llama.h location when built outside of root directory (#3179)

abetlen committed 1 year ago

Verified 76164fe2

ci : Cloud-V for RISC-V builds (#3160)

alitariq4589 committed 1 year ago

Verified c2ab6fe6

llama : remove mtest (#3177)

rbur0425 committed 1 year ago

Verified 2d770505

llama : make quantize example up to 2.7x faster (#3115)

cebtenzzre committed 1 year ago

Verified 98311c42

flake : allow $out/include to already exist (#3175)

jneem committed 1 year ago

Verified feea179e

cmake : compile ggml-rocm with -fpic when building shared library (#3158)

abetlen committed 1 year ago

Verified 769266a5

flake : include llama.h in nix output (#3159)

AsbjornOlling committed 1 year ago

Verified cf8238e7

make : fix clang++ detection, move some definitions to CPPFLAGS (#3155)

cebtenzzre committed 1 year ago

Verified 4b8560e7

CI: add FreeBSD & simplify CUDA windows (#3053)

alonfaraj committed 1 year ago

Verified 83a53b75

falcon : use stated vocab size (#2914)

akawrykow committed 1 year ago

Verified 5c872dbc

cmake : add relocatable Llama package (#2960)

bandoti committed 1 year ago

Verified 990a5e22

docker : add gpu image CI builds (#3103)

canardleteer committed 1 year ago

Verified 980ab41a

gguf-py : support identity operation in TensorNameMap (#3095)

KerfuffleV2 committed 1 year ago

Verified e3940841

feature : support Baichuan serial models (#3009)

jameswu2014 committed 1 year ago

Verified 4c8643dd

speculative : add heuristic algorithm (#3006)

leng-yue committed 1 year ago

Verified 35f73049

whisper : tokenizer fix + re-enable tokenizer test for LLaMa (#3096)

goerch committed 1 year ago

Verified 71ca2fad

cmake : add a compiler flag check for FP16 format (#3086)

RossComputerGuy committed 1 year ago

Verified 1b6c650d

CUDA: mul_mat_q RDNA2 tunings (#2910)

JohannesGaessler committed 1 year ago

Verified 0a5eebb4

Older