zhangyiss/mlx
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx.git synced 2025-12-16 01:49:05 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

Improve names of quantization arguments (#235)

Browse Source 
* Change the default quantization group_size to 64
* Rename groups to group_size and width to bits
This commit is contained in:
Angelos Katharopoulos
2023-12-20 16:53:53 -08:00
committed by GitHub
parent 57fe918cf8
commit b3916cbf2b
11 changed files with 184 additions and 180 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
mlx/backend/metal/quantized.cpp
View File

@@ -58,7 +58,7 @@ void QuantizedMatmul::eval_gpu(const std::vector<array>& inputs, array& out) {
if (B == 1) {
std::ostringstream kname;
kname << "qmv_" << (w_transposed ? "n_" : "t_") << type_to_name(out)
<< "_groups_" << groups_ << "_width_" << width_;
<< "_gs_" << group_size_ << "_b_" << bits_;
// Encode and dispatch kernel
auto compute_encoder = d.get_command_encoder(s.index);
@@ -87,7 +87,7 @@ void QuantizedMatmul::eval_gpu(const std::vector<array>& inputs, array& out) {
else {
std::ostringstream kname;
kname << "qmm_" << (w_transposed ? "t_" : "n_") << type_to_name(out)
<< "_groups_" << groups_ << "_width_" << width_;
<< "_gs_" << group_size_ << "_b_" << bits_;
// Encode and dispatch kernel
auto compute_encoder = d.get_command_encoder(s.index);