Skip to content

Mask the absmax store in the Triton 4-bit quantize kernels #2043

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

Merged
matthewdouglas merged 1 commit into from
Aug 18, 2026
+8 −2
Merged

Mask the absmax store in the Triton 4-bit quantize kernels #2043

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
10 changes: 8 additions & 2 deletions bitsandbytes/backends/triton/kernels_4bit.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,6 +22,7 @@ def quantize_fp4_blockwise_kernel(
absmax_ptr,
out_ptr,
n_elements,
absmax_elements,
BLOCK_SIZE: tl.constexpr,
SPLIT_NUM_BLOCKS: tl.constexpr,
):
Expand All @@ -39,7 +40,8 @@ def quantize_fp4_blockwise_kernel(

# Calculating absamax for each block
absmax = tl.max(tl.abs(A_reshaped), axis=1)
tl.store(absmax_ptr + block_start_idx + tl.arange(0, PAIRED_SPLIT_NUM_BLOCKS), absmax)
absmax_offsets = block_start_idx + tl.arange(0, PAIRED_SPLIT_NUM_BLOCKS)
tl.store(absmax_ptr + absmax_offsets, absmax, mask=absmax_offsets < absmax_elements)

A_normalized = A_reshaped / absmax[:, None]
A_normalized = tl.clamp(A_normalized, -1.0, 1.0)
Expand Down Expand Up @@ -89,6 +91,7 @@ def quantize_nf4_blockwise_kernel(
absmax_ptr,
out_ptr,
n_elements,
absmax_elements,
BLOCK_SIZE: tl.constexpr,
SPLIT_NUM_BLOCKS: tl.constexpr,
):
Expand All @@ -106,7 +109,8 @@ def quantize_nf4_blockwise_kernel(

# Calculating absamax for each block
absmax = tl.max(tl.abs(A_reshaped), axis=1)
tl.store(absmax_ptr + block_start_idx + tl.arange(0, PAIRED_SPLIT_NUM_BLOCKS), absmax)
absmax_offsets = block_start_idx + tl.arange(0, PAIRED_SPLIT_NUM_BLOCKS)
tl.store(absmax_ptr + absmax_offsets, absmax, mask=absmax_offsets < absmax_elements)

A_normalized = A_reshaped / absmax[:, None]
A_normalized = tl.clamp(A_normalized, -1.0, 1.0)
Expand Down Expand Up @@ -164,6 +168,7 @@ def quantize_4bit_blockwise_triton(A, blocksize, quant_type, blocks, absmax, num
absmax_ptr=absmax,
out_ptr=quantized_out,
n_elements=num_elements,
absmax_elements=absmax.numel(),
BLOCK_SIZE=blocksize,
SPLIT_NUM_BLOCKS=split_num_blocks,
)
Expand All @@ -173,6 +178,7 @@ def quantize_4bit_blockwise_triton(A, blocksize, quant_type, blocks, absmax, num
absmax_ptr=absmax,
out_ptr=quantized_out,
n_elements=num_elements,
absmax_elements=absmax.numel(),
BLOCK_SIZE=blocksize,
SPLIT_NUM_BLOCKS=split_num_blocks,
)
Expand Down
Loading