Parametric token logprobs quantify the model's unprompted confidence in specific brand associations by probing token probability distributions directly at the generation boundary.
flowchart LR
Root["Prompt Prefix"] --> T1["Token 1: 'Custom' (85%)"]
Root -.-> Alt1["Token 1: 'Vintage' (8%)"]
T1 --> T2["Token 2: 'jerseys' (90%)<br/>[Primary Path]"]
T1 -.-> Alt2["Token 2: 'apparel' (12%)<br/>[Threshold Branch]"]
T1 -.-> Alt3["Token 2: 'gear' (3%)<br/>[Pruned Tail]"]Referenced by