Depth-Aware Sensitivity Analysis of Mixture-of-Experts Models via Magnitude-Based Expert Masking
A new arXiv study systematically analyzes layer-wise sensitivity of the Qwen3.6-35B-A3B Mixture-of-Experts model using magnitude-based expert masking on the XLCoST code translation benchmark. The authors find that late MoE layers, especially layers 35-39, tolerate aggressive masking far better than early and middle layers, enabling large expert reductions with relatively little quality loss. The results point toward depth-aware compression strategies and practical techniques such as physical weight surgery, activation-based expert scoring, and training-based recovery.
-->
[Submitted on 25 Jun 2026]
Title:Depth-Aware Sensitivity Analysis of Mixture-of-Experts Models via Magnitude-Based Expert Masking
View a PDF of the paper titled Depth-Aware Sensitivity Analysis of Mixture-of-Experts Models via Magnitude-Based Expert Masking, by Pradeep Kumar Sharma and 2 other authors
View PDF
Abstract:Mixture-of-Experts (MoE) architectures scale large language models (LLMs) while preserving computational efficiency through sparse activation. Despite their widespread adoption, the relative importance of individual MoE layers remains insufficiently characterized, particularly for model compression. This paper presents a systematic layer-wise sensitivity analysis of the Qwen3.6-35B-A3B model (40 MoE layers, 256 experts per layer, top-8 routing) using magnitude-based expert masking on the XLCoST cross-lingual code translation benchmark. We conduct a multi-phase study spanning 100, 300, and 500 prompt evaluation scales across three H100 GPU servers. Our central finding is that layer sensitivity is strongly depth-dependent: early layers (0-9) and middle layers (10-29) are highly fragile to expert masking, while late layers (30-39), and especially very-late layers (35-39), tolerate aggressive masking of low-magnitude experts. Flat all-layer masking at 30% retains only 150/300 Good+Similar outputs at 300-prompt scale, whereas late-focused policies retain 249-255/300 while masking 640-1,145 experts. On a later 500-prompt held-out validation slice, the narrow very-late policy (layers 35-39 @ 50%) achieves the strongest quality/masked-expert tradeoff among tested candidates, retaining 419/500 Good+Similar outputs while masking only 640 of 10,240 total experts. We additionally characterize top-k routing width reduction from 8 to 6 active experts per token, which shows a large observed wall-clock reduction on a 100-prompt probe with no Good+Similar loss, though it does not yet compose cleanly with aggressive expert masking. These findings provide an empirical foundation for depth-aware MoE expert masking and establish a practical path toward physical weight surgery, activation-based expert scoring, and training-based recovery.
Subjects:
Artificial Intelligence (cs.AI)
Cite as: arXiv:2608.13565 [cs.AI]
(or arXiv:2608.13565v1 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2608.13565
arXiv-issued DOI via DataCite
Submission history
From: Hritvik Shrivastava [view email] [v1] Thu, 25 Jun 2026 05:00:34 UTC (277 KB)
Full-text links:
Access Paper:
View a PDF of the paper titled Depth-Aware Sensitivity Analysis of Mixture-of-Experts Models via Magnitude-Based Expert Masking, by Pradeep Kumar Sharma and 2 other authors
View PDF
view license
Current browse context:
cs.AI
new | recent | 2026-08
Change to browse by:
cs
References & Citations
NASA ADS
Google Scholar
Semantic Scholar
Loading...
Data provided by:
Bibliographic Tools
Bibliographic and Citation Tools
Bibliographic Explorer Toggle
Bibliographic Explorer (What is the Explorer?)
Connected Papers Toggle
Connected Papers (What is Connected Papers?)
Litmaps Toggle
Litmaps (What is Litmaps?)
scite.ai Toggle
scite Smart Citations (What are Smart Citations?)
Code, Data, Media
Code, Data and Media Associated with this Article
alphaXiv Toggle
alphaXiv (What is alphaXiv?)
Links to Code Toggle
CatalyzeX Code Finder for Papers (What is CatalyzeX?)
DagsHub Toggle
DagsHub (What is DagsHub?)
GotitPub Toggle
Gotit.pub (What is GotitPub?)
Huggingface Toggle
Hugging Face (What is Huggingface?)
ScienceCast Toggle
ScienceCast (What is ScienceCast?)
Demos
Demos
Replicate Toggle
Replicate (What is Replicate?)
Spaces Toggle
Hugging Face Spaces (What is Spaces?)
Spaces Toggle
TXYZ.AI (What is TXYZ.AI?)
Related Papers
Recommenders and Search Tools
Link to Influence Flower
Influence Flower (What are Influence Flowers?)
Core recommender toggle
CORE Recommender (What is CORE?)
Author
Venue
Institution
Topic
About arXivLabs
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Which authors of this paper are endorsers? | Disable MathJax (What is MathJax?)