MWOP: Modality-aware Width-wise Operation Pruning for Efficient MLLMs¶
L1 core-mechanism diagnostic. Executes independent path and channel masks; claims neither kernel speedup nor model-level accuracy.
Paper information¶
| Field | Details |
|---|---|
| Paper | MWOP: Modality-aware Width-wise Operation Pruning for Efficient MLLMs |
| Organization | No first-author affiliation is listed on the paper's first page. |
| First public release | 2026-10-01 (arXiv v1) |
| Authors' code | No public author implementation was found as of 2026-10-03. |
| Adapter / method | mwop |
| Local reproduction code | src/auto_research/foundation_latest_20261003.py |
Original paper¶
Motivation and method¶
Prune V2V, T2V, and T2T paths separately inside the same attention head and select distinct FFN channels for visual and text processing. Re-estimate FFN importance after attention pruning.
flowchart LR
I[Public input and current state] --> M[mwop mechanism]
M --> A[Auditable intermediates]
A --> O[Output or state update]Key figure from the paper¶
Figure 2: Overview of the paper's method and its key components. From the original paper; copyright remains with the authors. Click the image for the source.
Core equation¶
\(I(w)\approx|w\partial L/\partial w|\), with independent top-k selection for each modality path.
Paper results¶
The paper reports accuracy retention and throughput improvements at higher compression ratios on several MLLMs.
Local reproduction¶
Local comparison: The baseline disables the mechanism or uses default state; the treatment executes the defining operator. L1 reports no formal relative improvement, and relative percentages are not applicable.
- Three-seed diagnostic
diagnostic_only=true, ineligible for formal capability ranking.
Reproduction boundaries¶
Independent path and channel masks run locally. No kernel speedup or model-level accuracy claim is made.