Skip to content

Gdn qad #2455

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

Draft
sychen52 wants to merge 3 commits into
base: main
Choose a base branch
from
Draft

Gdn qad #2455

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
1bc6573
Vendor the flash-linear-attention chunked GatedDeltaNet kernels
sychen52 Sep 16, 2026
590b934
Fake-quantize the GatedDeltaNet recurrent state to FP8 inside the chu…
sychen52 Sep 17, 2026
13c7e24
Add an optional w_quantizer to the vendored chunked GatedDeltaNet kernel
sychen52 Sep 17, 2026
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
2 changes: 2 additions & 0 deletions .pre-commit-config.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -107,6 +107,8 @@ repos:
modelopt/onnx/quantization/ort_patching.py|
modelopt/torch/_deploy/utils/onnx_utils.py|
modelopt/torch/export/transformer_engine.py|
modelopt/torch/kernels/quantization/linear_attention/fla_chunk_delta_h.py|
modelopt/torch/kernels/quantization/linear_attention/fla_chunk_gated_delta_rule.py|
modelopt/torch/puzzletron/anymodel/models/gpt_oss/gpt_oss_pruned_to_mxfp4.py|
modelopt/torch/quantization/export_onnx.py|
modelopt/torch/quantization/plugins/attention.py|
Expand Down
1 change: 1 addition & 0 deletions LICENSE
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -250,6 +250,7 @@ the following copyright holders, licensed under the MIT License:
Copyright (c) 2023 DeepSeek
Copyright (c) 2025 sgl-project
Copyright (c) 2026 The DeepSpec Authors
Copyright (c) 2023-2026 Songlin Yang, Yu Zhang, Zhiyuan Li

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
Expand Down
24 changes: 24 additions & 0 deletions modelopt/torch/kernels/quantization/linear_attention/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,24 @@
# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

"""Linear-attention kernels for quantization.

``fla_chunk_delta_h.py`` and ``fla_chunk_gated_delta_rule.py`` are adapted copies of the chunked
GatedDeltaNet kernels of `flash-linear-attention <https://github.com/fla-org/flash-linear-attention>`_
(``fla.ops.common.chunk_delta_h`` and ``fla.ops.gated_delta_rule.chunk``) that can fake-quantize the
recurrent state carried between chunks to FP8 (``state_qdq``). They still import the surrounding
fla operators, so ``flash-linear-attention`` (v0.5.1 or newer) and Triton must be installed to use
them. This package initializer does not import the kernels, so importing it needs neither.
"""
Loading
Loading