Code

Q2 and Q3 quantization#1004

Code

Closed

sw wants to merge 4 commits intoggml-org:masterggml-org/llama.cpp:masterfrom sw:q2q3sw/llama.cpp:q2q3Copy head branch name to clipboard

+452-17

Commits

Commits on Apr 20, 2023

• Q2 and Q3 quantization

  sw

  committed
  6fc51a8

  Copy full SHA for 6fc51a8

• Q2 AVX2: do two blocks at a time, by @slaren

  sw

  committed
  c29ab90

  Copy full SHA for c29ab90

• More AVX2 optimizations

  sw

  committed
  8c90a86

  Copy full SHA for 8c90a86

• Faster q3_0 implementation, using two planes, by @pubby

  

  pubby

  authored and

  sw

  committed
  7aa501c

  Copy full SHA for 7aa501c