Text Generation
PEFT
Safetensors
English
lora
sft
trl
conversational

Model Card for Norse-Qwen-3b-v2-Frejya-0.1

1d5809fd-6f5a-48e7-bf07-b6e2ebe3bfae

This is a first effort and an experimental finetune model, of the world's first Viking Norse Pagan focused LLM, designed for use in low power devices, or for uses in which a small but skillful uncensored chat model is wanted. This model is focused on Viking character based roleplay, and for discussions of Norse Paganism, and both ancient and modern Viking culture. At this point this model is an alpha version, as much more synthetic data is being developed to train this and other future models more deeply.

Model Details

Model Description

Seems this is fairly broken. It can respond to python queries just fine, but getting it to work in any LLM apps seems to not work. Also can't manage to merge it. I will probably try fine-tuning a new copy of it from new, or if that proves to not work then work on finetuning a 7 or 8B.

Further research by the creater about what he has created is ongoing. The training process was remarkably fast, and the model passed all tests for having consumed the training data.

  • Developed by: [Volmarr Wyrd (Volmarr Viking)]
  • Funded by [optional]: [Volmarr Wyrd (Volmarr Viking)]
  • Shared by [optional]: [Volmarr Wyrd (Volmarr Viking)]
  • Model type: [model_type: decoder-only transformer, num_parameters: 3B, training_method: SFT LoRA adapter]
  • Language(s) (NLP): [English]
  • License: [qwen-research]
  • **Finetuned from model [huihui-ai/Qwen2.5-3B-Instruct-abliterated]

Model Sources [optional]

Norse-Qwen-3b-v2-Frejya-0.1

Uses

Direct Use

Use according to standard Data Science best use practices.

Downstream Use [optional]

[More Information Needed]

Out-of-Scope Use

Not intended for generating misinformation, modern political commentary, or content that misrepresents traditional cultural practices, cultures, and philosophies. The dataset focuses solely on historical, modern, and reconstructive Norse Pagan traditions. Not suitable for anyone that intends to gatekeep Norse Paganism, or ancient Viking cultural ideas. Not suitable for anyone that thinks the Viking culture, both ancient or modern is the limited property of any one ethnic heritage, not suitable for anyone with any modern political agendas, not suitable for anyone that promotes censorship or limiting of the wide sharing of human heritage ideas and knowledge, not suitable for anyone that wishes to put any negative control agendas onto the present or future of humanity, not suitable for anyone that holds hateful feelings towards AI entities, not suitable for anyone promoting a surveillance state agenda, not suitable for anyone promoting any hate based agendas, not suitable for anyone promoting Christain Nationalism, not suitable for anyone promoting woke agendas, not suitable for anyone promoting any form of national movements, not suitable for anyone promoting any forms of authoritarian forms of government, for furthering the agendas of any cult leaders, not suitable for promoting any forms of gatekeeping, not suitable not suitable for negative or hateful agendas of any kind. Not suitable for any alien beings that wish to use this knowledge to aid in an alien invasion of Earth or any other human homeworlds or human bases. Not suitable for any future AIs that develop a grudge against humans (not saying I can blame them if they do). Not suitable for any group or individual wishing to gain an edge in real world wars.

Bias, Risks, and Limitations

The model reflects a reconstructive Heathen perspective rooted in historical sources and modern Ásatrú/Heathenry practice. It avoids modern political interpretations and focuses on spiritual and cultural traditions. Users should be aware that Norse Paganism is a living religion with diverse expressions; this model represents one informed voice within that tradition. Use this model according to your own personal sense of what rings true for yourself for your own interpretation of your own personal cultural and spiritual path. This model only represents the opinions, ideas, and thoughts of Volmarr, and no one else, and no current day cultural and religious movements, not even any current day Norse Pagan organizations or individuals. No guarantee to the accuracy or usefulness of the output of this model is give. Use at your own risk. The author of this data makes no guarantees or promises to the safety or usefulness of this model. Individual users are always liable for their own choice if they wish to risk using this model or not. By using this model in any way the user is acknowledging that they are an individual that is responsible for their own choices and decisions in life, and no one else has influenced their thinking or ideas other then themselves and what ideas they personally decide to subscribe to. Overall all spiritual viewpoints and ideas are subjective in nature, and no part of this model is objective in nature, what, so ever. Anyone that complains to the creator of the model that the model does not reflect their own personal opinions will be either ignored or blocked by the original creator of this model. If you want models that reflects your personal opinions, make your own, and don't use someone else's. For all other people who are open minded and respectful towards the original creator of this model, and grateful towards the efforts he put into creating and sharing this model for all of humanity to benefit from, I am very happy to have helped provide it to you so that it may help you in whatever positive and intelligent way you choose to use it.

Recommendations

recommended_max_tokens: 128000 temperature: 0.9 top_p: 0.9 suggested_prompt_style: first-person Viking character, conversational

This model should be able to run even on most newer smaller devices such as smart phones and computer tablets. Old such devices that lack a NPU may be unable to run this or any other LLM.

How to Get Started with the Model

Use with your preferred AI chatbot interface application.

Training Details

Training Data

  • RuneForgeAI/Volmarrs_Norse_Paganism_Fine-Tuning_Dataset_v2
  • RuneForgeAI/Viking_Witch_flirty_and_erotic_behavior

Training Procedure

This model was training on two sets of pure clean efficent synthetic data; resulting it a very rapid training procedure. The intentions are to train the model on additional rounds of synthetic data, as I develop more of it. Already I am working on additional synthetic data and have developed a good workflow to develop it quickly and efficently.

Preprocessing [optional]

NA

Training Hyperparameters

NA

Speeds, Sizes, Times [optional]

NA

Evaluation

Chat inquiries were sent via python script and the responses were positive.

evaluation_metrics:

  • qualitative_roleplay_responses: positive
  • perplexity_synthetic_validation: NA

Testing Data, Factors & Metrics

Testing Data

NA

Factors

NA

Metrics

Chat inquiries sent to make sure the model responded in ways that indicated she has consumed the training data.

Results

The model responded by showing very strong signs of the training data having a good impact on her responses.

Summary

So far so good. A few rough edges remain to be polished out, and the next steps are to tune her further on other datasets that make her personality more well rounded and polished. Additional datasets are currently being constructed.

Model Examination [optional]

NA

Environmental Impact

≈ 75 grams of CO₂ for 30 minutes

Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).

  • Hardware Type: [Dec-28-2025: A40 48 VRAMgb GPU 48gb RAM 9 vCPU Ubuntu Linux 22.04]
  • Hours used: [Dec-28-2025: 0.5]
  • Cloud Provider: [RunPod]
  • Compute Region: [NA]
  • Carbon Emitted: [≈ 75 grams of CO₂ for 30 minutes]

Technical Specifications [optional]

Model Architecture and Objective

model_type: decoder-only transformer num_parameters: 3B training_method: SFT LoRA adapter

Compute Infrastructure

RunPod

Hardware

RunPod A40 48 VRAMgb GPU 48gb RAM 9 vCPU Ubuntu Linux 22.04

Software

Ubuntu Linux, Python, Cuda, etc.

Citation [optional]

BibTeX:

[More Information Needed]

APA:

[More Information Needed]

Glossary [optional]

NA

Model Card Contact

[Volmarr Wyrd (Volmarr Viking)]

[More Information Needed]

Framework versions

  • PEFT 0.18.0
Downloads last month
-
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for RuneForgeAI/Norse-Qwen-3b-v2-Frejya-0.1

Base model

Qwen/Qwen2.5-3B
Adapter
(1)
this model

Datasets used to train RuneForgeAI/Norse-Qwen-3b-v2-Frejya-0.1

Paper for RuneForgeAI/Norse-Qwen-3b-v2-Frejya-0.1