SVG Generation
The capability of an AI model to describe visual content by generating SVG (Scalable Vector Graphics) code. This requires the model to translate visual intent into precise geometric coordinates and path instructions, and serves as a method for evaluating a model's genuine visual understanding by assessing how accurately it can produce structured graphical output.
Timeline (last 90 days)
SVG generation models output graphics as code token by token, requiring a forward pass at each step, which results in slow drawing speed.
In SVG tasks, drafter errors in speculative decoding are very rare and exhibit a clustered distribution, and the distances over which accuracy is maintained within a block are much deeper than in natural language.
All Facts (2)
SVG generation models output graphics as code token by token, requiring a forward pass at each step, which results in slow drawing speed.
50%UnverifiedIn SVG tasks, drafter errors in speculative decoding are very rare and exhibit a clustered distribution, and the distances over which accuracy is maintained within a block are much deeper than in natural language.
50%