llama.cpp
ggml : PoC for normalizing weights for better quantization packing
#2434

Open

Commits

ggml : poc for normalizing weights for better quantization (metal)

ggerganov committed 2 years ago
ggml : use less ggml_mul tasks when src0 rows are few

ggerganov committed 2 years ago
cuda : poc for norm quants (only -b 1 works)

ggerganov committed 2 years ago

Loading