ggml-org/ggml

Pull Requests Commits

ggerganov committed 2 years ago

Verified 46e6c0fd

ggerganov committed 2 years ago

Verified a21da9fe

ggerganov committed 2 years ago

Verified 3ad84136

sync : llama.cpp

ggerganov committed 2 years ago

Verified 6c1ce0bd

metal : update ggml-metal.m from llama.cpp

ggerganov committed 2 years ago

Verified 8c745c3c

ggml : add IQ2 to test-backend-ops + refactoring (llama/4990)

ggerganov committed 2 years ago

Verified e055f5f1

imatrix : offload to GPU support (llama/4957)

ggerganov committed 2 years ago

Verified 2467de43

backend : add eval callback (llama/4935)

ggerganov committed 2 years ago

Verified 85a03e94

metal : create autorelease pool during library build (llama/4970)

ggerganov committed 2 years ago

Verified 612dabb7

ggml : importance matrix support for legacy quants (llama/4969)

ikawrakow committed 2 years ago

Verified 006c363a

metal : log `recommendedMaxWorkingSetSize` on iOS 16+ (llama/4936)

azarovalex committed 2 years ago

Verified fd8267ab

ggml : introduce GGML_CALL function annotation (llama/4850)

jart committed 2 years ago

Verified b3bcd556

cuda : fix dequantize kernel names (llama/4938)

ggerganov committed 2 years ago

Verified 4624d4ae

CUDA: faster dequantize kernels for Q4_0 and Q4_1 (llama/4938)

ikawrakow committed 2 years ago

Verified a00a9f28

Add ability to use importance matrix for all k-quants (llama/4930)

ikawrakow committed 2 years ago

Verified 327eb4b6

sync : whisper.cpp

ggerganov committed 2 years ago

Verified b306d6e9

sync : llama.cpp

ggerganov committed 2 years ago

Verified bca51b52

metal : correctly set SIMD support flags on iOS (llama/4923)

azarovalex committed 2 years ago

Verified 15bfd124

2-bit quantizations (llama/4897)

ikawrakow committed 2 years ago

Verified 82f6044a

sync : whisper.cpp

ggerganov committed 2 years ago

Verified 1890780d

whisper : load the model into multiple buffers of max size 1GB (whisper/1763)

ggerganov committed 2 years ago

Verified 3fb4dcb6

sync : llama.cpp

ggerganov committed 2 years ago

Verified 1553a05a

examples : adapt to metal API

ggerganov committed 2 years ago

Verified aea44652

ggml: cache sin/cos for RoPE (llama/4908)

JohannesGaessler committed 2 years ago

Verified 04621342

metal : remove old API (llama/4919)

ggerganov committed 2 years ago

Verified 2065272c

metal : disable log for loaded kernels (llama/4794)

ggerganov committed 2 years ago

Verified a19f18ef

gguf : fix potential infinite for-loop (llama/4600)

texmex76 committed 2 years ago

Verified 8cc38282

metal : refactor kernel loading code (llama/4794)

ggerganov committed 2 years ago

Verified 4a5e71ec

CUDA: faster q8_0 -> f16 dequantization (llama/4895)

JohannesGaessler committed 2 years ago

Verified bcdb75e3

sync : whisper.cpp

ggerganov committed 2 years ago

Verified 400c07f0

Older