Commits

History for

Commits on May 27, 2024

update HIP_UMA ggerganov#7399 (ggerganov#7414 )

Djip007
and
slaren
committedMay 27, 2024
Allow multiple copy function pointers for CUDA graph kernel param updates (ggerganov#7565 )
agray3
committedMay 27, 2024

Commits on May 19, 2024

cuda : clear error after buffer allocation failure (ggerganov#7376 )
slaren
committedMay 19, 2024

Commits on May 18, 2024

Capture CUDA logging output (ggerganov#7298 )

fraxy-v
and
slaren
committedMay 18, 2024

Commits on May 15, 2024

Avoid unnecessarily disabling CUDA graphs (ggerganov#7302 )
agray3
committedMay 15, 2024

Commits on May 12, 2024

CUDA: add FP32 FlashAttention vector kernel (ggerganov#7188 )
JohannesGaessler
committedMay 12, 2024

Commits on May 11, 2024

feat: implemented sigmoid function (ggml/806)

justcho5
authored and
ggerganov
committedMay 11, 2024
ggml : full ALiBi support (ggerganov#7192 )
ggerganov
committedMay 11, 2024

Commits on May 8, 2024

Introduction of CUDA Graphs to LLama.cpp (ggerganov#6766 )

agray3
and
slaren
committedMay 8, 2024

Commits on May 6, 2024

Add an option to build without CUDA VMM (ggerganov#7067 )
WilliamTambellini
committedMay 6, 2024

Commits on Apr 30, 2024

ggml : add Flash Attention (ggerganov#5021 )

committedApr 30, 2024

Commits on Apr 18, 2024

ggml : group all experts in a single ggml_mul_mat_id (ggerganov#6505 )

slaren
and
ggerganov
committedApr 18, 2024

Commits on Apr 13, 2024

CUDA: fix matrix multiplication logic for tests (ggerganov#6667 )
JohannesGaessler
committedApr 13, 2024

Commits on Apr 9, 2024

llama : add Command R Plus support (ggerganov#6491 )

committedApr 9, 2024

Commits on Apr 7, 2024

ggml: bypass code incompatible with CUDA < 11.1 (whisper/2020)

primenko-v
authored and
ggerganov
committedApr 7, 2024

Commits on Apr 3, 2024

ggml : mul_mat_id use the same tensor for all the experts (ggerganov#6387 )

slaren
and
ggerganov
committedApr 3, 2024

Commits on Mar 26, 2024

llama : greatly reduce output buffer memory usage (ggerganov#6122 )

compilade
and
ggerganov
committedMar 26, 2024
IQ1_M: 1.75 bpw quantization (ggerganov#6302 )

ikawrakow
and
Kawrakow
committedMar 26, 2024

Commits on Mar 25, 2024

cuda : refactor into multiple files (ggerganov#6269 )
slaren
committedMar 25, 2024

Commits on Mar 22, 2024

cuda : add LLAMA_CUDA_NO_PEER_COPY to workaround broken ROCm p2p copy (ggerganov#6208 )
slaren
committedMar 22, 2024

Commits on Mar 21, 2024

cuda : disable host register by default (ggerganov#6206 )
slaren
committedMar 21, 2024
cuda : fix LLAMA_CUDA_F16 build (ggerganov#6197 )
slaren
committedMar 21, 2024
Add ability to use Q5_0, Q5_1, and IQ4_NL for quantized K cache (ggerganov#6183 )

ikawrakow
and
Kawrakow
committedMar 21, 2024
cuda : fix conflict with std::swap (ggerganov#6186 )
slaren
committedMar 21, 2024

Commits on Mar 20, 2024

cuda : print the returned error when CUDA initialization fails (ggerganov#6185 )
slaren
committedMar 20, 2024
cuda : refactor to remove global resources (ggerganov#6170 )
slaren
committedMar 20, 2024

Commits on Mar 18, 2024

backend : offload large batches to GPU (ggerganov#6083 )

slaren
and
JohannesGaessler
committedMar 18, 2024

Commits on Mar 15, 2024

cuda : disable unused cudaLaunchHostFunc code (ggerganov#6078 )
slaren
committedMar 15, 2024

Commits on Mar 13, 2024

llama : add pipeline parallelism support (ggerganov#6017 )

committedMar 13, 2024

Commits on Mar 12, 2024

ggml : reuse quantum structs across backends (ggerganov#5943 )
ggerganov
committedMar 12, 2024

Commits on Mar 11, 2024

1.5 bit: we can do even better (ggerganov#5999 )

ikawrakow
and
Kawrakow
committedMar 11, 2024
Better 1.5 bit quantization (ggerganov#5971 )

ikawrakow
and
Kawrakow
committedMar 11, 2024

Commits on Mar 9, 2024

ggml : add ggml-common.h to deduplicate shared code (ggerganov#5940 )
ggerganov
committedMar 9, 2024

Commits on Mar 4, 2024

ggml : introduce ggml_status (ggml/750)

committedMar 4, 2024
add some new ops, fix some operators and add batch operations to certain operators. (ggml/747)

committedMar 4, 2024