ggml-org/llama.cpp

Pull Requests Commits

ggerganov committed 2 years ago

Verified 2f689dee

common : don't do warm-up with more than n_batch tokens (close #3058)

ggerganov committed 2 years ago

Verified efac2d46

metal : restore original F16 mat-vec multiplication

ggerganov committed 2 years ago

Verified 78337967

metal : put warning in kernel_norm to not combine the loops

ggerganov committed 2 years ago

Verified ed92c3d4

metal : fix kernel_norm

ggerganov committed 2 years ago

Verified 5e1c4089

ggml : posixify madvise and pagesize (#3037)

przemoc committed 2 years ago

Verified fec2fb19

k-quants : fix zero-weight guard in Q6_K (ref #3040)

ggerganov committed 2 years ago

Verified 178b1850

convert-llama-ggml-to-gguf: Try to handle files older than GGJTv3 (#3023)

KerfuffleV2 committed 2 years ago

Verified ea2c85d5

build : add LLAMA_METAL_NDEBUG flag (#3033)

cebtenzzre committed 2 years ago

Verified 9912b9ef

make : use new flag variables for recent changes (#3019)

cebtenzzre committed 2 years ago

Verified 9e202315

examples : replace fprintf to stdout with printf (#3017)

cebtenzzre committed 2 years ago

Verified de2fe892

convert: fix convert.py not working with int filename_stem (#3028)

Green-Sky committed 2 years ago

Verified c9c3220c

Guard against all weights in a super-block being zero (#3010)

ikawrakow committed 2 years ago

Verified d59bd970

llama : update logic for number of threads when using BLAS

ggerganov committed 2 years ago

Verified 35938ee3

speculative : add grammar support (#2991)

ggerganov committed 2 years ago

Verified 92177210

ggerganov committed 2 years ago

Verified 2ba85c86

build : on Mac OS enable Metal by default (#2901)

ggerganov committed 2 years ago

Verified e36ecdcc

ggml-opencl : store GPU buffer in ggml_tensor::extra (#2994)

slaren committed 2 years ago

Verified bd33e5ab

llama-bench : make cpp file non-executable (#2999)

cebtenzzre committed 2 years ago

Verified 31035681

make : add speculative example (#3003)

leng-yue committed 2 years ago

Verified 5b8530d8

server : add a subtle loading animation to the edit box (#2466)

akx committed 2 years ago

Verified e4386f41

2x faster (rms) norm cuda kernels (3.7% e2e improvement) (#2985)

li-plus committed 2 years ago

Verified 35195689

ggml-alloc : use virtual memory for measurement (#2973)

slaren committed 2 years ago

Verified cf9b0848

speculative : PoC for speeding-up inference via speculative sampling (#2926)

ggerganov committed 2 years ago

Verified 47068e51

perplexity : fix ETA by warming up the model with an empty run

ggerganov committed 2 years ago

Verified 8f429fa5

gguf(python): Fix special vocab handling when id < 0 (#2984)

KerfuffleV2 committed 2 years ago

Verified 6519e9c9

metal : restore 363f0bf and fix reduce in F16_F32 kernels (#2986)

ggerganov committed 2 years ago

Verified b7f2aa9e

cov : disable comment in PRs (#2989)

alonfaraj committed 2 years ago

Verified 73a12a63

llama : fix bpe tokenize from byte (#2889)

opparco committed 2 years ago

Verified 37301347

metal : revert 6af0bab until we fix it

ggerganov committed 2 years ago

Verified d9151e6f

Older