ngxson/llama.cpp

Pull Requests Commits

`server`: inject date_string in llama 3.x template + fix date for firefunction v2 (#12802)

ochafik committed 234 days ago

Verified aa48e373

kv-cache : fix out-of-bounds view during reserve graph (#13547)

ggerganov committed 234 days ago

Verified e3a9421b

arm64: optimize q6_k_q8_k kernel with i8mm (#13519)

cyb70289 committed 234 days ago

Verified 5ab5d5fb

`common`: add partial regex support (#12808)

ochafik committed 234 days ago

Verified 3198405e

editorconfig : fix trailing whitespace from #13542 (#13546)

CISC committed 235 days ago

Verified f5170c1d

fix: crash when calling `llama_state_get_size` on a context without a KV cache (#13542)

giladgd committed 235 days ago

Verified 017f10b5

CUDA: fix crash on large batch size for quant. MoE (#13537)

JohannesGaessler committed 235 days ago

Verified 4696d567

llama : fix quantize with dl backends (#13539)

slaren committed 235 days ago

Verified b7d26720

CUDA: faster Deepseek FA, add Turing support (#13435)

JohannesGaessler committed 235 days ago

Verified 6da34fa2

fix: Move build_inp_pos to the top of the graph section for build_granite (#13538)

gabe-l-hart committed 235 days ago

Verified 5e7d95e2

server : passthrough the /models endpoint during loading (#13535)

ggerganov committed 235 days ago

Verified 05317443

server : fix cache_tokens bug with no cache_prompt (#13533)

ngxson committed 235 days ago

Verified 360a9c98

cmake: simplify vulkan shader test logic (#13263)

bandoti committed 235 days ago

Verified 09d13d94

vulkan: KHR_coopmat flash attention (#13506)

jeffbolznv committed 235 days ago

Verified 24e86cae

webui : use fflate for more deterministic gzip compress (#13525)

ngxson committed 235 days ago

Verified bb1681fb

webui: Allow pasting file from clipboard (#13526)

luca020400 committed 235 days ago

Verified d486dd3e

docs: Update link to ggml-org in multimodal.md (#13513)

ddpasa committed 235 days ago

Verified 21ca987f

scripts : fix compare-llama-bench.py show parameter (#13514)

CISC committed 235 days ago

Verified be1d4a13

vulkan: workaround FA compile failures on macos (#13517)

jeffbolznv committed 235 days ago

Verified ab3971f2

quantize : improve tensor-type pattern matching (#13033)

EAddario committed 236 days ago

Verified e5c834f7

clip : clip.h become private API (⚠️ breaking change) (#13510)

ngxson committed 236 days ago

Verified 71bdbdb5

metal : use FA-vec kernel up to batch size 20 (#13496)

ggerganov committed 236 days ago

Verified f0995d28

metal : optimize multi-sequence FA vec kernel (#13493)

ggerganov committed 236 days ago

Verified c252e0c4

ggml-cpu: Update KleidiAI to v1.6 and fix include directives (#13509)

eddnjjn committed 236 days ago

Verified 4f711afe

batched-bench : fix pp batch contents (#13492)

ggerganov committed 236 days ago

Verified b89d605a

mtmd : remove libllava, remove clip-quantize-cli (⚠️ breaking change) (#13460)

ngxson committed 236 days ago

Verified b4726345

scripts : support arbitrary input file formats in compare-llama-bench.py (#13455)

CISC committed 236 days ago

Verified bf793711

model : Granite MoE shared (#13269)

gabe-l-hart committed 236 days ago

Verified d590cd4c

ggerganov committed 236 days ago

1e2809bc

llama-bench : add defrag-thold, check for invalid ranges (#13487)

slaren committed 236 days ago

Verified cf0a43bb

Newer Older