Skip to content

Apply linear rope_scaling in model builder for Neutts/nano #2142

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
kunal-vaishnavi merged 4 commits into from
May 18, 2026
Merged

Apply linear rope_scaling in model builder for Neutts/nano #2142

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
0d181ee
Apply Hugging Face linear rope_scaling in model builder
VishalX Apr 9, 2026
004a30a
Added Modification copyright
VishalX Apr 22, 2026
c0d1568
Remove duplicated copyright comment
VishalX May 8, 2026
43b3fca
Remove print statement
VishalX May 18, 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
13 changes: 12 additions & 1 deletion src/python/py/models/builders/base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,7 +3,7 @@
# Licensed under the MIT License. See License.txt in the project root for
# license information.
#
# Copyright (C) [2026] Advanced Micro Devices, Inc. All rights reserved.
# Modifications Copyright (C) 2026 Advanced Micro Devices, Inc. All rights reserved.
# Portions of this file consist of AI generated content.
# --------------------------------------------------------------------------
from __future__ import annotations
Expand Down Expand Up @@ -466,6 +466,17 @@ def make_rope_init(self, config):
"ntk_beta": beta_fast,
}

elif (
config.rope_scaling.get("rope_type", config.rope_scaling.get("type")) == "linear"
and "factor" in config.rope_scaling
):
# Hugging Face: modeling_rope_utils._compute_linear_scaling_rope_parameters — inv_freq /= factor
# Equivalent to inv_freq = 1 / (factor * theta ** (i / dim)) in make_rotary_embedding_caches_from_scratch.
factor = float(config.rope_scaling["factor"])
if factor <= 0:
raise ValueError(f"rope_scaling.factor must be positive for linear RoPE scaling, got {factor}")
self.rope_attrs["rescale_factors"] = factor

elif "mrope_section" in config.rope_scaling:
# For models that use MRoPE (e.g. Qwen 2.5 VL, Qwen 3 VL)
self.rope_attrs["mrope"] = {
Expand Down