ggml-org/llama.cpp

Pull Requests Commits

server : re-enable swa speculative decoding

ggerganov committed 1 year ago

Verified 36fce982

context : simplify output counting logic during decode (#14142)

ggerganov committed 1 year ago

Verified f6e1a7aa

batch : remove logits_all flag (#14141)

ggerganov committed 1 year ago

Verified c3ee46fa

cmake : handle whitepsaces in path during metal build (#14126)

ggerganov committed 1 year ago

Verified e2c0b6e4

kv-cache : fix split_equal handling in unified implementation (#14130)

ggerganov committed 1 year ago

Verified 95965069

context : round n_tokens to next multiple of n_seqs when reserving (#14140)

compilade committed 1 year ago

Verified a20b2b05

common: fix issue with regex_escape routine on windows (#14133)

mtmcp committed 1 year ago

Verified 2e89f76b

Implement GGML_CPU_ALL_VARIANTS for ARM (#14080)

ckastner committed 1 year ago

Verified 532802f9

chore : clean up relative source dir paths (#14128)

CISC committed 1 year ago

Verified d4e0d95c

tests : add test-tokenizers-repo (#14017)

CISC committed 1 year ago

Verified cc66a7f7

vulkan: Better thread-safety for command pools/buffers (#14116)

jeffbolznv committed 1 year ago

Verified bd248d4d

webui: Wrap long numbers instead of infinite horizontal scroll (#14062)

am17an committed 1 year ago

Verified 7781e5fe

kv-cache : relax SWA masking condition (#14119)

ggerganov committed 1 year ago

Verified 89a184fa

server : pass default --keep argument (#14120)

MightyAlex200 committed 1 year ago

Verified 2baf0772

kv-cache : add LLAMA_KV_CACHE_DEBUG environment variable (#14121)

ggerganov committed 1 year ago

Verified 7ae29321

vulkan: Track descriptor pools/sets per-context (#14109)

jeffbolznv committed 1 year ago

Verified 1f7d50b2

opencl: add `mul_mv_id_q4_0_f32_8x_flat` (#14003)

lhez committed 1 year ago

Verified 4c763c8d

kv-cache : avoid modifying recurrent cells when setting inputs (#13834)

compilade committed 1 year ago

Verified dad5c443

convert : fix duplicate key DeepSeek-R1 conversion error (#14103)

CISC committed 1 year ago

Verified 55f6b9fa

llama : support GEGLU for jina-bert-v2 (#14090)

CISC committed 1 year ago

Verified 3678b838

vulkan: force device 0 in CI (#14106)

jeffbolznv committed 1 year ago

Verified 652b70e6

Fixed spec timings to: accepted/tested instead of accepted/drafted (#14104)

jukofyork committed 1 year ago

Verified 3a12db23

ggerganov committed 1 year ago

ae92c185

ggml : fix weak alias win32 (whisper/0)

ggerganov committed 1 year ago

b7ce1ad1

Vulkan: Don't default to CPU device (like llvmpipe), even if no other device is available, to allow fallback to CPU backend (#14099)

0cc4m committed 1 year ago

Verified 97340b4c

rpc : nicer error messages for RPC server crash (#14076)

isaac-mcfadyen committed 1 year ago

Verified 2bb04670

ggerganov committed 1 year ago

b8e2194e

Add in-build ggml::ggml ALIAS library (ggml/1260)

dg0yt committed 1 year ago

1a3b5e80

metal : use less stack memory in FA kernel (#14088)

ggerganov committed 1 year ago

Verified 1f63e75f

kv-cache : fix shift and defrag logic (#14081)

ggerganov committed 1 year ago

Verified 40cbf571

Older