Character Model R&D
We study how characters can retain identity, memory, and relationships across long interactions, so each response can build on what came before.
What matters is not only what a character says, but whether it can remain the same character over time
Character Consistency
Maintain persona, voice, values, and behavioral boundaries across long interactions while reducing drift as context grows.
Relationship Continuity
Carry shared experience and relationship changes into later decisions, so interaction has a past instead of resetting every turn.
Narrative Causality
Keep events, choices, and consequences traceably connected so character behavior still holds over longer time horizons.
Using a small model to test the value of focused training
qwen3-4b-tylogiorm
qwen3-4b-tylogiorm is a post-trained roleplay model based on Qwen/Qwen3-4B. It is designed for users who care more about character consistency, distinctive voice, and narrative continuity than general assistant coverage.
Its goal is not to become a universal frontier assistant. It is narrower and more practical: make a small model genuinely strong at immersive roleplay.
In the archived SillyBench tests, this specialization produced a large jump over the untuned Qwen3-4B base model and reached a level that competes with much larger reference models on roleplay-specific evaluation.
View model recordsJudge stability through long interaction, not a single-turn impression
SillyBench observes persona drift, emotional discontinuity, repetitive language, and narrative collapse across many turns, grounding training direction in results that can be identified and compared.
Observe whether the character drifts as context grows
Check whether persona and shared experience continue to shape later interaction
Identify emotional breaks, repetition, and narrative collapse
浙公网安备33011002019019号