llama.cpp
ggml : add Q8_0 quantization format (rename the old one to Q8_1) (ARM NEON)
#1179

Merged

ggml : add Q8_0 quantization format (rename the old one to Q8_1) (ARM NEON) #1179

ggerganov merged 10 commits into master from q8_0

ggml : add Q8_0 quantization format (rename the old one to Q8_1)

f83c321c

tests : fix test-quantize-fns

79cfdf5e

ggml : finalize Q8_0 implementation

d8bf7207

ggml : use q4_0_q8_0 and q4_2_q8_0

6496b79e

ggml : fix Q8_0 dot product bug (ARM)

88618ab7

ggml : Q8_0 unroll x2

6e0f0b6f

ggerganov added generation quality

ggerganov assigned

ggerganov 3 years ago

ggml : fix bug - using wrong block type

46fc696d

ggerganov force pushed to 46fc696d 3 years ago

ggml : extend quantize_fns_t with "vec_dot_type"

91bfa51d

sw commented on 2023-04-25

ggml : fix Q8_0 to use 255 values out of 256

4ddb983a

sw commented on 2023-04-25

ggml : fix assert using wrong QK4_2 instead of QK4_3

e8c37317

ggerganov merged 7a32fcb3 into master 3 years ago

ggerganov deleted the q8_0 branch 3 years ago

mofosyne added Tensor Encoding Scheme

mofosyne added Review Complexity : High

Reviewers

Assignees

ggerganov

Labels

generation quality Review Complexity : High Tensor Encoding Scheme

Milestone

No milestone