ggerganov/llama.cpp

Pull Requests Commits

llama : make starcoder graph build more consistent with others

ggerganov committed 1 year ago

Verified 92a4f868

metal : fix out-of-bounds access in soft_max kernels

ggerganov committed 1 year ago

Verified f82328ab

cleanup useless code

wsxiaoys committed 1 year ago

6c353dc7

working in cpu, metal buggy

wsxiaoys committed 1 year ago

a1cf66ea

wsxiaoys committed 1 year ago

101c5787

add input embeddings handling

wsxiaoys committed 1 year ago

8bc76a22

store mqa directly

wsxiaoys committed 1 year ago

ab13d071

fix vram calculation for starcoder

wsxiaoys committed 1 year ago

4420cff6

wsxiaoys committed 1 year ago

dac31da4

fix head count kv

wsxiaoys committed 1 year ago

0be15e16

properly load all starcoder params

wsxiaoys committed 1 year ago

77c7ec17

set n_positions to max_positioin_embeddings

wsxiaoys committed 1 year ago

26836119

add max_position_embeddings

wsxiaoys committed 1 year ago

a17ef397

load starcoder weight

wsxiaoys committed 1 year ago

57f064d7

set head_count_kv = 1

wsxiaoys committed 1 year ago

166a259f

add LLM_ARCH_STARCODER to llama.cpp

wsxiaoys committed 1 year ago

7298c37e

fix ffn_down name

wsxiaoys committed 1 year ago

7e0a843b

convert MQA to MHA

wsxiaoys committed 1 year ago

76d32cca

support convert starcoder weights to gguf

wsxiaoys committed 1 year ago

eb7f0eba

add placeholder of starcoder in gguf / llama.cpp

wsxiaoys committed 1 year ago

0c5d4d87

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

Older