3 and 6 bit quantization (#1613)

* Support 3 and 6 bit quantization
2025-12-16 01:49:05 +08:00 · 2024-11-22 10:22:13 -08:00
parent 0c5eea226b
commit c79f6a4a8c
12 changed files with 633 additions and 419 deletions
--- a/mlx/ops.cpp
+++ b/mlx/ops.cpp
@@ -3683,7 +3683,7 @@ std::tuple<array, array, array> quantize(
    int group_size /* = 64 */,
    int bits /* = 4 */,
    StreamOrDevice s /* = {} */) {
-  return fast::affine_quantize(w, group_size, bits);
+  return fast::affine_quantize(w, group_size, bits, s);
 }

 array dequantize(