---

base_model: google/gemma-4-31B-it library_name: peft tags:

• lora
• peft
• gemma
• japanese
• roleplay
• text-generation language:
• ja

---

COLK RP v0.3 — Japanese Roleplay LoRA

Overview

COLK RP v0.3 is an experimental LoRA adapter for google/gemma-4-31B-it, developed to explore Japanese-language creative roleplay and character-driven conversations.

The goal is to improve roleplay writing style and dialogue quality while preserving the capabilities of the underlying model.

This is an early experimental release, not a fully validated production model.

Intended Use

• Japanese-language fictional roleplay
• Character-driven dialogue
• Creative storytelling
• Experimentation with RP writing styles

The adapter is designed to be used with its corresponding base model.

Model Information

Property	Value
Base model	google/gemma-4-31B-it
Adapter type	PEFT LoRA
LoRA rank	16
LoRA alpha	32
Version	v0.3
Primary language	Japanese

Files

• adapter_config.json — LoRA configuration
• adapter_model.safetensors — Trained adapter weights

This repository does not include the base model weights.

Training

This adapter was fine-tuned using a small, curated dataset focused on Japanese fictional roleplay.

The training aimed to explore improvements in natural dialogue, expressive narration, and character-oriented writing.

Testing and Limitations

The adapter has been tested locally using a quantized GGUF-based setup.

The original adapter has not yet been thoroughly evaluated in a non-quantized inference environment.

Performance may vary depending on the inference backend, context length, prompt structure, and generation settings.

The adapter does not guarantee consistent characterization or accurate long-context instruction following.

Status

Experimental — further evaluation and refinement are planned.

License

License and redistribution terms are not yet specified. Please contact the repository owner before redistributing the adapter files.