BWhat changed
Distributed groups of visual tokens across aligned language/vision transformer layers through residual injection, allowing richer visual inputs without placing all tokens in the language input sequence.
Scientific scope
Distributed groups of visual tokens across aligned language/vision transformer layers through residual injection, allowing richer visual inputs without placing all tokens in the language input sequence. Residual fusion and visual tokenization are inherited. Savings concern context placement and measured compute, not zero processing cost for all additional tokens. Original multiscale-token construction is distinct from later Qwen3VL intermediate-feature adaptation.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Residual fusion and visual tokenization are inherited. Savings concern context placement and measured compute, not zero processing cost for all additional tokens. Original multiscale-token construction is distinct from later Qwen3VL intermediate-feature adaptation.
Contributors
- Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction