ggml-org/llama.cpp

Pull Requests Commits

cuda : add flash_attn kernel (wip)

ggerganov committed 1 year ago

Verified b957b8f5

cuda : fix soft_max to use correct mask size

ggerganov committed 1 year ago

Verified 2e460137

ggml : fix ggml_soft_max mask requirement

ggerganov committed 1 year ago

Verified 910b15bb

ggml : switch to padded F16 mask for ggml_soft_max, ggml_flash_attn_ext

ggerganov committed 1 year ago

Verified 8ad92dc1

Merge branch 'master' into gg/flash-attn

ggerganov committed 1 year ago

Verified 2ddc9bbe

llama : reorder build_orion() at correct place (#5118)

ggerganov committed 1 year ago

Verified d3bac7d5

llama : remove LLAMA_MAX_DEVICES and LLAMA_SUPPORTS_GPU_OFFLOAD (#5240)

ggerganov committed 1 year ago

Verified 5cb04dbc

metal : add im2col F32 dst support (#5132)

ggerganov committed 1 year ago

Verified efb7bdbb

llava : add MobileVLM support (#5132)

JidongZhang-THU committed 1 year ago

Verified 15606309

format license text, restore apache license by legal suggestion (#5233)

NeoZhangJianyu committed 1 year ago

Verified b2b9f025

ggml : limit n_threads to the max n_tasks (#5238)

slaren committed 1 year ago

Verified dabcc5b4

Vulkan Fixes (#5223)

0cc4m committed 1 year ago

Verified f8e9140c

Fix typos of IQ2_XXS and IQ3_XXS in llama.cpp (#5231)

ymcui committed 1 year ago

Verified d62520eb

support SYCL backend windows build (#5208)

NeoZhangJianyu committed 1 year ago

Verified 01684139

kompute : llama-bench support and ggml_cpu_has_kompute() (#5226)

cebtenzzre committed 1 year ago

Verified e8dc55d0

Merge branch 'master' into gg/flash-attn

ggerganov committed 1 year ago

Verified 3d03bcb7

ggerganov committed 1 year ago

Verified 78df5527

metal : fix array initialization

ggerganov committed 1 year ago

Verified d073e4f9

Revert "server : change deps.sh xxd files to string literals (#5221)"

ggerganov committed 1 year ago

Verified e0085fdf

server : fix context shift (#5195)

ggerganov committed 1 year ago

Verified e6f291d1

server : change deps.sh xxd files to string literals (#5221)

jboero committed 1 year ago

Verified 4003be0e

ggml : fix IQ3_XXS on Metal (#5219)

ikawrakow committed 1 year ago

Verified fea4fd4b

sync : ggml (#0)

ggerganov committed 1 year ago

Verified 8f8ddfcf

gguf : fix comparison (ggml/715)

ggerganov committed 1 year ago

Verified 6fb50ebb

`ggml_cuda_cpy` support for 4d tensors and float16->float32 upcasting (ggml/686)

balisujohn committed 1 year ago

Verified 625a699b

gguf : add input validation, prevent integer overflows (ggml/709)

ggerganov committed 1 year ago

Verified a4b07c05

ci : fix yolo URLs + fix metal capture (ggml/712)

ggerganov committed 1 year ago

Verified 549a1e6c

metal : add debug capture backend function (ggml/694)

jmousseau committed 1 year ago

Verified 5f14ee0b

Faster AVX2 dot product for IQ2_XS (#5187)

ikawrakow committed 1 year ago

Verified 8e14e3dd

SOTA 3-bit quants (#5196)

ikawrakow committed 1 year ago

Verified f4d7e549

Older