Introduces a multimodal lens for tracing how vision-language models represent visual concepts and selectively access them in response to questions. Counterfactual tests examine how those concepts affect later reasoning; the work does not claim models are conscious.
Runs shared Transformer layers multiple times to increase per-token computation while keeping the model compact, targeting local multimodal and agentic workloads.
Provides quantized GGUF builds of Inkling 2 Mini for local text-and-image inference, with file sizes from about 7.76 to 22.87 GB. Runtime support for its looped transformer architecture may vary.