Commits

Commits on May 28, 2024

ggml : fix another case of quants nans (llama/7387)

slaren
authored and
ggerganov
committedMay 28, 2024
cuda : clear error after buffer allocation failure (llama/7376)

slaren
authored and
ggerganov
committedMay 28, 2024
ggml : fix quants nans when all the group weights are very close to zero (llama/7313)

slaren
authored and
ggerganov
committedMay 28, 2024
ggml : tag ggml_tensor::backend as deprecated (llama/7290)

slaren
authored and
ggerganov
committedMay 28, 2024

Commits on May 11, 2024

add basic tensor data validation function (llama/6884)

slaren
authored and
ggerganov
committedMay 11, 2024
gguf : fix mismatch between alloc and free functions (llama/6929)

slaren
authored and
ggerganov
committedMay 11, 2024
ggml : group all experts in a single ggml_mul_mat_id (llama/6505)

slaren
and
ggerganov
committedMay 11, 2024
metal : unify mul_mv_id kernels (llama/6556)

slaren
authored and
ggerganov
committedMay 11, 2024

Commits on Apr 6, 2024

ggml : mul_mat_id use the same tensor for all the experts (llama/6387)

slaren
and
ggerganov
committedApr 6, 2024
ggml : fix bounds checking of zero size views (llama/6347)

slaren
authored and
ggerganov
committedApr 6, 2024

Commits on Mar 27, 2024

cuda : rename build flag to LLAMA_CUDA (llama/6299)

slaren
authored and
ggerganov
committedMar 27, 2024
cuda : refactor into multiple files (llama/6269)

slaren
authored and
ggerganov
committedMar 27, 2024
cuda : add LLAMA_CUDA_NO_PEER_COPY to workaround broken ROCm p2p copy (llama/6208)

slaren
authored and
ggerganov
committedMar 27, 2024
cuda : disable host register by default (llama/6206)

slaren
authored and
ggerganov
committedMar 27, 2024
cuda : fix LLAMA_CUDA_F16 build (llama/6197)

slaren
authored and
ggerganov
committedMar 27, 2024
cuda : fix conflict with std::swap (llama/6186)

slaren
authored and
ggerganov
committedMar 27, 2024
cuda : print the returned error when CUDA initialization fails (llama/6185)

slaren
authored and
ggerganov
committedMar 27, 2024
cuda : refactor to remove global resources (llama/6170)

slaren
authored and
ggerganov
committedMar 27, 2024
backend : set max split inputs to GGML_MAX_SRC (llama/6137)

slaren
authored and
ggerganov
committedMar 27, 2024
backend : offload large batches to GPU (llama/6083)

authored and
ggerganov
committedMar 27, 2024
cuda : disable unused cudaLaunchHostFunc code (llama/6078)

slaren
authored and
ggerganov
committedMar 27, 2024

Commits on Mar 18, 2024

alloc : fix allocation data of pre-allocated leafs

slaren
authored and
ggerganov
committedMar 18, 2024

Commits on Mar 14, 2024

update examples and tests

slaren
authored and
ggerganov
committedMar 14, 2024
llama : add pipeline parallelism support (llama/6017)

committedMar 14, 2024
test-backend-ops : skip CPU backend by default (llama/6028)

slaren
authored and
ggerganov
committedMar 14, 2024

Commits on Mar 7, 2024

magika : fix ggml_backend_graph_compute status check (closes ggerganov#760 )
slaren
committedMar 7, 2024

Commits on Mar 4, 2024

cuda : fix data race in soft max (llama/5853)

slaren
authored and
ggerganov
committedMar 4, 2024

Commits on Feb 25, 2024

add google magika inference example (ggerganov#748 )
slaren
committedFeb 25, 2024

Commits on Feb 21, 2024

cuda : ignore peer access already enabled errors (llama/5597)

slaren
authored and
ggerganov
committedFeb 21, 2024

Commits on Feb 19, 2024

cuda, metal : fix nans in soft_max (llama/5574)

slaren
and
ggerganov
committedFeb 19, 2024
cuda : print message when initialization fails (llama/5512)

slaren
authored and
ggerganov
committedFeb 19, 2024

Commits on Feb 12, 2024

ggml-alloc : allocate all leafs as if they were inputs (ggerganov#731 )
slaren
committedFeb 12, 2024

Commits on Feb 11, 2024

ggml-alloc : v3 (ggerganov#727 )
slaren
committedFeb 11, 2024

Commits on Feb 10, 2024

cuda : fix LLAMA_CUDA_F16 (llama/5262)

slaren
authored and
ggerganov
committedFeb 10, 2024
ggml : limit n_threads to the max n_tasks (llama/5238)

slaren
authored and
ggerganov
committedFeb 10, 2024