ggerganov/llama.cpp

Pull Requests Commits

[Fix] Reenable server embedding endpoint (#1937)

SlyEcho committed 2 years ago

Verified 20568fe6

ggml : fix bug in LBFGS optimizer (found by ggml tests)

ggerganov committed 2 years ago

Verified 18b35625

llama : use aligned memory during ggml_init call from loading saved sessions (#1934)

l3utterfly committed 2 years ago

Verified ba4e85a8

cmake : fix trailing whitespaces

ggerganov committed 2 years ago

Verified 23fc5c21

llama : only use Q6_K for output weights if tensor size is multiple of 256 (#1932)

ikawrakow committed 2 years ago

Verified cb40dfca

cuda : faster k-quants on older GPUs (#1930)

ikawrakow committed 2 years ago

Verified ca7c3f4d

ggml : sync latest ggml repo (#1924)

ggerganov committed 2 years ago

Verified b97ca431

cmake : fix build shared ggml when CUDA is enabled (#1929)

howard0su committed 2 years ago

Verified 1e3abfce

Convert vector to f16 for dequantize mul mat vec (#1913)

JohannesGaessler committed 2 years ago

Verified 16b9cd19

Added tokens per second to info prints (#1928)

JohannesGaessler committed 2 years ago

Verified b24c3049

Fixed incorrectly applying RMS norm twice (#1925)

JohannesGaessler committed 2 years ago

Verified 0ede372a

ggml : fix bug in ggml_compute_forward_add_q_f32 (#1918)

l3utterfly committed 2 years ago

Verified 8596af42

readme : update Android build instructions (#1922)

mikeyang01 committed 2 years ago

Verified e1886cf4

llama : prevent usage of k-quants when tensor size is not a multiple of 256 (#1921)

ikawrakow committed 2 years ago

Verified 8ab8ba62

examples : fix examples/metal (#1920)

ikawrakow committed 2 years ago

Verified 90cc59d6

metal : handle buffers larger than device's maxBufferLength (#1826)

ggerganov committed 2 years ago

Verified ce2c7d72

cmake : add CUDA_ARCHITECTURES to new target ggml_static (#1917)

howard0su committed 2 years ago

Verified 57cd6946

make : do not print help for simple example

ggerganov committed 2 years ago

Verified b2416493

minor : warning fixes

ggerganov committed 2 years ago

Verified 4f9c43e3

Only one CUDA stream per device for async compute (#1898)

JohannesGaessler committed 2 years ago

Verified 2c9380dd

llama : fix kv_cache `n` init (close #1903)

ggerganov committed 2 years ago

Verified 051e1b0e

make : update for latest Arch (#1701)

daniandtheweb committed 2 years ago

Verified 86c75718

ggml : fix warnings under MSVC (#1908)

howard0su committed 2 years ago

Verified 3d59ec59

metal : add norm, cpy f16->f16, alibi kernels (#1823)

apage43 committed 2 years ago

Verified 0711a5f6

exposed modules so that they can be invoked by nix run github:ggerganov/llama.cpp#server etc (#1863)

faezs committed 2 years ago

Verified fc45a81b

Server Example Refactor and Improvements (#1570)

digiwombat committed 2 years ago

Verified 794db3e7

hooks : setting up flake8 and pre-commit hooks (#1681)

jpodivin committed 2 years ago

Verified 5ddf7ea1

readme : alternative way to build for Android with CLBlast. (#1828)

gustrd committed 2 years ago

Verified bac19927

Allow cmake to build ggml as a library (#1896)

KerfuffleV2 committed 2 years ago

Verified b4c6f46f

train : get raw text instead of page with html (#1905)

Davidy22 committed 2 years ago

Verified 92f20d99

Newer Older