Darksp33d/loreweaver-rp-v1

LoreWeaver RP v1

LoreWeaver RP v1 is a parameter-efficient role-play and interactive-fiction fine-tune of the Qwen3.6-35B-A3B family. It is intended for vivid scene writing, sustained character voice, dialogue, and long-form narrative continuation.

This repository contains a LoRA adapter, not a standalone copy of the base model. A compatible Qwen3.6-35B-A3B base checkpoint is required for inference.

Model summary

Field	Value
Developer	Darksp33d
Model	Darksp33d/loreweaver-rp-v1
Release date	July 19, 2026
Task	Autoregressive text generation
Specialization	Role-play, interactive fiction, creative writing, and character dialogue
Model family	Qwen3.6-35B-A3B
Public architectural parent	Qwen/Qwen3.6-35B-A3B
Adapter format	PEFT LoRA in Safetensors format
Adapter parameters	13,762,560
Adapter size	Approximately 55.1 MB
Context declared by tokenizer/config	262,144 tokens
Primary documented language	English
License	Not yet declared for this adapter; the public Qwen parent is Apache-2.0

The base architecture is a mixture-of-experts model with 40 text layers, 256 experts, and 8 experts selected per token. The public parent contains approximately 35.95B parameters. The adapter repository contains text-model configuration and does not establish that vision behavior was trained, retained, or evaluated; this model card therefore documents LoreWeaver RP v1 as a text-generation adapter.

Intended uses

LoreWeaver RP v1 is designed for:

• interactive story continuation;
• in-character conversation and dialogue;
• character-driven role-play;
• scene narration and descriptive prose;
• worldbuilding grounded in supplied lore or session context;
• branching narrative applications in which the user directs the story.

The model is most useful inside an application that manages character definitions, world lore, memory, recent dialogue, and output constraints. It can also be used directly with a clear system prompt and conversation history.

Out-of-scope uses

This adapter was not developed or evaluated for:

• factual question answering or retrieval;
• medical, legal, financial, or other high-stakes decisions;
• autonomous real-world actions;
• identity verification or claims about real people;
• code generation;
• image, audio, or video understanding;
• guaranteed policy enforcement or content moderation.

Do not treat generated narrative as factual, and do not use the model to impersonate a real person without their consent.

Prompt format

Use the tokenizer and chat template shipped with the adapter repository. The template accepts system, user, assistant, and tool roles.

LoreWeaver RP v1 is intended to return finished narrative output rather than visible reasoning. For runtimes that support the included template arguments, set:

enable_thinking=False

A minimal conversation looks like:

messages = [
    {
        "role": "system",
        "content": (
            "Write immersive interactive fiction. Maintain character voice, "
            "respect established lore, and never decide the player's actions."
        ),
    },
    {
        "role": "user",
        "content": "I push open the observatory door and call for Mara.",
    },