ngxson/llama.cpp

Pull Requests Commits

readme : add rpc backend (#12842)

ggerganov committed 320 days ago

Verified 47277d6d

CANN: Support Opt CONV_TRANSPOSE_1D and ELU (#12786)

noemotiovon committed 320 days ago

Verified 6e1c4ceb

vulkan: In coopmat2 mmq, load q4_k/q5_k scales through shared memory (#12833)

jeffbolznv committed 320 days ago

Verified 0090950f

vulkan: Use fp16 for the flash attention P*V multiplication (#12783)

jeffbolznv committed 320 days ago

Verified 7ecd780b

cuda : add f32 to bf16 copy op (#12806)

CISC committed 321 days ago

Verified 7538246e

llava: improve clip_ctx destructor to not memleak load_image_size (#12834)

mattjcly committed 321 days ago

Verified b32efad2

llama : fix FA when KV cache is not used (i.e. embeddings) (#12825)

ggerganov committed 321 days ago

Verified a19b5cef

server : fix thread.join() on exit (#12831)

ngxson committed 321 days ago

Verified 78a1ba0a

llava: add more helper functions to check projector types in clip context (#12824)

dm4 committed 321 days ago

Verified 2dabf759

arg : Including limits file on AIX (#12822)

mehendarkarprajwal committed 321 days ago

Verified 1d343b40

server : webui : Improve Chat Input with Auto-Sizing Textarea (#12785)

characharm committed 321 days ago

Verified 8ca6e1c3

Revert "sycl:remove redundant memcopy in function ggml_backend_sycl_buffer_set_tensor" (#12812)

NeoZhangJianyu committed 321 days ago

Verified 656babd6

gguf-py : support lazy tensor splitting (#12809)

compilade committed 321 days ago

Verified a226bc7a

llama : Support llama 4 text-only (#12791)

ngxson committed 322 days ago

Verified 1466621e

opencl: better identify Adreno GPU (#12760)

lhez committed 322 days ago

Verified 82974011

hellaswag: display estimated score confidence interval (#12797)

stduhpf committed 322 days ago

Verified 4ccea213

cuda : fix HIP and MUSA BF16 (#0)

ggerganov committed 322 days ago

1a1ab7e7

ggerganov committed 322 days ago

a4e46e28

ggml : simplify Arm fp16 CPU logic (ggml/1177)

ggerganov committed 322 days ago

ff067dbc

CUDA: don't convert BF16 weights to FP32 (ggml/1174)

CISC committed 322 days ago

36ca8b36

cpu: move all the operators into a separate c++ file (except mul_mat) (ggml/1167)

cmdr2 committed 322 days ago

995083e4

sycl: remove redundant memcopy in function ggml_backend_sycl_buffer_set_tensor (#12734)

jeffzhou2000 committed 322 days ago

Verified 518a0148

ci : no curl on ggml-ci (#12796)

ngxson committed 322 days ago

Verified e391d3ee

cmake : enable curl by default (#12761)

ngxson committed 322 days ago

Verified bd3f59f8

CANN: fix typo in ggml-cann (#12733)

jeffzhou2000 committed 322 days ago

Verified 52b3d71f

CANN: Refactor to reduce duplicate code (#12731)

hipudding committed 322 days ago

Verified d0d5b223

musa: fix compilation warnings in mp_22/31 (#12780)

yeahdongcn committed 323 days ago

Verified 916c83bf

vulkan: fix NaN issue in flash attention shader (#12776)

jeffbolznv committed 323 days ago

Verified 0c74b043

vulkan: Use unclamped loads for flash attention mask (#12720)

jeffbolznv committed 323 days ago

Verified 80b717d4

Vulkan: Tune Vulkan mmq int dot shader for performance (#12767)

0cc4m committed 324 days ago

Verified 6bf28f01

Newer Older