Skip to content

feat(quant): thread a mode argument through gather_qmm #358

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
sergey-scherbina wants to merge 1 commit into
base: main
Choose a base branch
from
Open

feat(quant): thread a mode argument through gather_qmm #358

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 15 additions & 2 deletions mlx-rs/src/ops/quantization.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -174,10 +174,12 @@ pub fn dequantize_device<'a>(
/// - `group_size`: The quantization group size (default: 64)
/// - `bits`: The number of bits per element (default: 4)
/// - `sorted_indices`: If true, indicates the indices are sorted (default: false)
/// - `mode`: Quantization mode - `"affine"` (scales + biases) or a microscaling
/// format such as `"mxfp4"` (scales only, no biases) (default: `"affine"`)
#[allow(clippy::too_many_arguments)]
#[generate_macro]
#[default_device]
pub fn gather_qmm_device<'b, 'lhs, 'rhs>(
pub fn gather_qmm_device<'a, 'b, 'lhs, 'rhs>(
x: impl AsRef<Array>,
w: impl AsRef<Array>,
scales: impl AsRef<Array>,
Expand All @@ -188,12 +190,19 @@ pub fn gather_qmm_device<'b, 'lhs, 'rhs>(
#[optional] group_size: impl Into<Option<i32>>,
#[optional] bits: impl Into<Option<i32>>,
#[optional] sorted_indices: impl Into<Option<bool>>,
#[optional] mode: impl Into<Option<&'a str>>,
#[optional] stream: impl AsRef<Stream>,
) -> Result<Array> {
let transpose = transpose.into().unwrap_or(true);
let group_size = optional_int(group_size.into(), DEFAULT_GROUP_SIZE);
let bits = optional_int(bits.into(), DEFAULT_BITS);
let sorted = sorted_indices.into().unwrap_or(false);
// `mode` selects the quantization scheme: `"affine"` (scales + biases) or a
// microscaling format such as `"mxfp4"` (scales only, no zero-point biases),
// matching `mx.gather_qmm(..., mode=...)`. Threaded through instead of
// hardcoding `DEFAULT_MODE` so MoE experts stored in mxfp4 work.
let mode_str = mode.into().unwrap_or("affine");
let mode_cstr = std::ffi::CString::new(mode_str).expect("Invalid mode string");

unsafe {
let biases_ptr = biases
Expand Down Expand Up @@ -221,7 +230,7 @@ pub fn gather_qmm_device<'b, 'lhs, 'rhs>(
transpose,
group_size,
bits,
DEFAULT_MODE.as_ptr(),
mode_cstr.as_ptr(),
sorted,
stream.as_ref().as_ptr(),
)
Expand Down Expand Up @@ -391,6 +400,8 @@ mod tests {
group_size,
bits,
None,
// mode: default "affine" (scales + biases)
None,
)
.unwrap();
assert!(
Expand All @@ -414,6 +425,8 @@ mod tests {
group_size,
bits,
None,
// mode: default "affine" (scales + biases)
None,
)
.unwrap();
assert!(
Expand Down