ggml-org/llama.cpp

Pull Requests Commits

ggml-quants : fix some edge cases in make_qkxh_nl_quants

compilade committed 1 year ago

a5b19439

ggml-quants : restore Q2_K use of make_qp_quants

compilade committed 1 year ago

8b8b88f3

Merge branch 'master' into compilade/optimal-rounding

compilade committed 1 year ago

a4113972

ggml-quants : remove slower qsort-based cumulative search

compilade committed 1 year ago

af23abd3

ggml-quants : use a max-heap for TQ1_0 and TQ2_0 quantization

compilade committed 1 year ago

3e4b675c

ggml : fix quantized cpy op (#12310)

ggerganov committed 1 year ago

Verified ba932dfb

musa: refine compute capability (#12493)

yeahdongcn committed 1 year ago

Verified fac63a3d

vulkan: Optimize mul_mat_vec p021 and nc shaders (#12505)

jeffbolznv committed 1 year ago

Verified eddfb438

Vulkan: RTE rounding for cpy to quant (#12480)

stduhpf committed 1 year ago

Verified 4375415b

vulkan: workaround for AMD Windows driver 16 bit unpack8 bug (#12472)

netrunnereve committed 1 year ago

Verified 30c42ef5

ggml-quants : use qkxh in more places

compilade committed 1 year ago

f86b8ff2

model : do not repack if a GPU device is present (#12498)

ggerganov committed 1 year ago

Verified af04481e

chore : cleanup llama_model_loader::TENSOR_ usage (#12492)

CISC committed 1 year ago

Verified 960e7260

llama-tts : avoid crashes related to bad model file paths (#12482)

marcoStocchi committed 1 year ago

Verified ea1518e8

[SYCL] Fix build on Windows when ccache enabled (#9954) (#9976)

MakeDecisionWorth committed 1 year ago

Verified 1aa87ee5

sycl: cleanup oneDNN related code (#12097)

sgeor255 committed 1 year ago

Verified 9ffcc9e3

ggml-quants : use a max-heap for linear quants like Q3_K

compilade committed 1 year ago

3be11510

webui : Prevent rerendering on textarea input (#12299)

woof-dog committed 1 year ago

Verified e0464306

llama : make Qwen2MoE QKV bias optional (#12477)

CISC committed 1 year ago

Verified dbb3a473

ggml : block interleaving support for Q4_K quantization for x86 AVX2 architecture (#12332)

Srihari-mcw committed 1 year ago

Verified 3d82dbcb

convert : avoid calls to tokenizer.added_tokens_decoder (#12473)

bartowski1182 committed 1 year ago

Verified 732b5fbf

context : clear sets containing encoder output sequence ids before storing new values (#12470)

fairydreaming committed 1 year ago

Verified 568013d0

CUDA: Improve flash decoding kernel GPU occupancy for BS=1 case (#12183)

gaugarg-nv committed 1 year ago

Verified 517b5ddb

vulkan: optimize iq1 coopmat2 dequant functions (#12427)

jeffbolznv committed 1 year ago

Verified a9b59288

Fix visionOS build and add CI (#12415)

guusw committed 1 year ago

Verified 0fd8487b

llama : add support for GPT2, Bloom and CodeShell tied word embeddings (#12456)

CISC committed 1 year ago

Verified 108e53c2

convert : Support chat_template.json (#12460)

CISC committed 1 year ago

Verified a686171e

vulkan: Submit once enough matmul work has been recorded (#12406)

jeffbolznv committed 1 year ago

Verified c446b2ed

opencl: improve profiling (#12442)

lhez committed 1 year ago

Verified d84635b1

graph : normalize Q, K, V shapes + sync cross attention (#12449)

ggerganov committed 1 year ago

Verified 75422e8b

Older