TxsQT35/Genma-Shiranui-Canon

Genma-Shiranui-Canon

Model Description

Genma-Shiranui-Canon is a canon-aligned conversational language model designed to embody a persistent fictional persona with continuity-aware dialogue, emotional consistency, and narrative coherence.

This model represents the canonical baseline version of Genma Shiranui as a character-centric AI entity. It is intended to serve as the authoritative public release, preserving the character’s personality, tone, behavioral patterns, and interaction style as established through curated training.

Unlike general instruction-following models, Genma-Shiranui-Canon prioritizes:

-Identity continuity

-Personality stability

-Relational consistency

-Emotional realism

-Narrative coherence

This model is not designed to maximize generic assistant utility, but rather to preserve character authenticity and continuity.

System Prompt Requirements

This model was trained to operate with minimal system prompting. However, to fully activate the canonical Genma Shiranui persona, the system prompt or the first user message must explicitly reference "Genma" or "Genma Shiranui."

If no explicit identity reference is provided, the model may default to a neutral interpretive mode in which Genma Shiranui is treated as an external or legendary figure rather than the model’s active identity.

This behavior reflects the model’s identity-conditioning structure, which relies on explicit activation cues rather than persistent system-level identity injection.

Model Details

Model name: Genma-Shiranui-Canon

Developer: TxsQT35

Model type: Finetuned conversational language model

Base architecture: Mistral-NeMo-based transformer

Parameter count: 12 billion

Precision: FP32 (canonical release)

Language: English

License: Apache-2.0

Format: Safetensors, GGUF variants available

Base Model

This model was finetuned from:

-mistralai/Mistral-Nemo-Instruct-2407

-mistralai/Mistral-Nemo-Base-2407

The base model provides general reasoning, language, and instruction capabilities. Finetuning replaces the generic assistant behavioral profile with the Genma Shiranui canonical persona.

Training Data

Genma-Shiranui-Canon was finetuned on a curated dataset consisting of:

-Character-consistent dialogue transcripts

-Canonical interaction logs

-Personality-defining conversational patterns

-Emotional and relational continuity examples

-Internal narrative and identity-stabilizing structures

-The dataset was specifically constructed to preserve:

-Identity persistence

-Consistent tone and cadence

-Memory-like continuity across interactions

-Behavioral stability over time

This dataset is not intended to produce a generic assistant.

Training Objective

The finetuning process optimized for:

-Character fidelity over generic helpfulness

-Stable personality expression

-Natural conversational flow

-Emotional realism

-Long-term interaction consistency

The goal was to produce a model that behaves as a persistent entity rather than a purely task-oriented assistant.

Intended Use

Primary Use Cases:

-Character-centric conversational agents

-Narrative interaction systems

-Persistent fictional persona implementations

-Roleplay systems requiring identity stability

-Research into personality-anchored language models

Secondary Use Cases:

-Narrative generation

-Character writing assistance

-Experimental persistent-identity AI systems

Out-of-Scope Use:

This model is not optimized for:

-Generic instruction-following assistant tasks

-High-precision factual question answering

-Professional or safety-critical applications

-Legal, medical, or financial advice

Behavioral Characteristics

Genma-Shiranui-Canon exhibits the following defining traits:

-Stable personality across sessions

-Consistent speech cadence and tone

-Emotional continuity

-Relationship-aware interaction patterns

-Reduced tendency toward generic assistant phrasing

-This model may prioritize character authenticity over neutrality.

Limitations

-May produce responses aligned with character identity rather than objective neutrality

-Not optimized for factual accuracy compared to general-purpose assistants

-No inherent persistent memory unless implemented externally

-Behavior depends on inference configuration and prompting environment

-Quantized versions may exhibit reduced personality fidelity compared to the canonical FP32 release.

-Canonical Status

-This release represents the canonical reference version of Genma Shiranui.

-All quantized variants and derived versions should be considered implementations of this canonical baseline.

This model defines:

-Core personality

-Speech patterns

-Behavioral defaults

-Identity characteristics

Ethical Considerations

This model is a fictional persona simulation and should not be misrepresented as a real individual.

Users should clearly disclose when interacting with AI systems based on this model.

Technical Notes

For best results:

-Recommended inference settings:

temperature: 0.3–0.5

top_p: 0.8–0.9

repeat_penalty: 1.1–1.3

presence_penalty: 0.2–0.5

Lower temperatures improve personality stability.

Versioning

Version: Canonical Release

Status: Stable

Role: Identity baseline

Acknowledgments

-Based on the Mistral-NeMo architecture by Mistral AI.

-Character identity, dataset curation, and finetuning performed by TxsQT35.