ggerganov/llama.cpp

Pull Requests Commits

CUDA: return -1 for nonexistent compiled arch (#15587)

JohannesGaessler committed 11 days ago

Verified 8f5afa94

metal : optimize FA vec for large sequences and BS <= 8 (#15566)

ggerganov committed 11 days ago

Verified b3964c1e

mtmd : support Kimi VL model (#15458)

ngxson committed 11 days ago

Verified 79a54622

context : print graph stats for memory-less contexts (#15586)

ggerganov committed 11 days ago

Verified 85cc1ae9

metal : improve `MUL_MAT_ID` (#15541)

ggerganov committed 11 days ago

Verified 1d8d83de

model : support MiniCPM-V 4.5 (#15575)

tc-mb committed 11 days ago

Verified c4e92390

gguf-py : remove erroneous FFN_GATE entry (#15583)

CISC committed 11 days ago

Verified 39842a7f

metal : remove contiguous assertion for src0 in IM2COL (#15577)

CISC committed 11 days ago

Verified 0fd90db5

Add a warning for special devices (#15563)

pt13762104 committed 11 days ago

Verified 4c37636b

vulkan: Remove splitting for mul_mat_id (#15568)

jeffbolznv committed 11 days ago

Verified 34bdbbd7

CUDA: Accelerate MXFP4 table lookup using `__byte_perm` (#15451)

Qeeweew committed 12 days ago

Verified 74f52f77

opencl: fix support ops condition for `rms_norm` (#15560)

lhez committed 12 days ago

Verified f7207b04

vulkan: fix min subgroup 16 condition for mmid subgroup optimization (#15565)

0cc4m committed 12 days ago

Verified 4d917cd4

tests: Generate unique input values for count_equal (#15487)

jeffbolznv committed 12 days ago

Verified 886b97a5

metal: fix regression when no metal devices are present (#15531)

booxter committed 12 days ago

Verified 111f8d06

CUDA: MoE helper in device code, better tile sizes (#15525)

JohannesGaessler committed 12 days ago

Verified 5eff6ec9

model-conversion : set pooling type to none in logits.cpp (#15564)

danbev committed 12 days ago

Verified dfd9b5f6

model-conversion : add model card template for embeddings [no ci] (#15557)

danbev committed 12 days ago

Verified 5a6bc6b1

batched-bench : fix unified KV cache handling + pp timing (#15562)

ggerganov committed 12 days ago

Verified 6b64f74b

convert : update Ernie 4.5 dense architecture name (#15555)

ownia committed 12 days ago

Verified 0d5a4702

metal : add FA kernels for HS=40 (#15559)

ggerganov committed 12 days ago

Verified b0ba31f5

convert : support interns1-mini (#15412)

RunningLeon committed 12 days ago

Verified 7da9fed0

CANN: ROPE cache sin/cos repeat (#15501)

noemotiovon committed 13 days ago

Verified c247d06f

vulkan: apply MUL_MAT_ID subgroup optimization to non-coopmat devices (#15524)

0cc4m committed 13 days ago

Verified 043fb27d

kv-cache : support layer reuse (#15504)

ggerganov committed 13 days ago

Verified b730706a

vulkan: Support FA with any multiple of 8 head sizes (#15537)

jeffbolznv committed 13 days ago

Verified c9a24fb9

vulkan: enable Conv2D for Apple after MoltenVK fixed the bug (#15526)

0cc4m committed 13 days ago

Verified a9c6ffcb

vulkan: workaround MoltenVK compile failure in multi_add (#15506)

jeffbolznv committed 13 days ago

Verified e78cf0d4

CUDA: fix half2 -> half conversion for HIP (#15529)

JohannesGaessler committed 14 days ago

Verified 710dfc46

vulkan: optimize rms_norm, and allow the work to spread across multiple SMs (#15281)

jeffbolznv committed 14 days ago

Verified 611f419c

Newer Older