ewald1976/g4-12b-it-trismegistus
The ewald1976/g4-12b-it-trismegistus is a 12 billion parameter instruction-tuned language model, fine-tuned by ewald1976 from the Gemma 4 12B Instruct base model. It specializes in esoteric, occult, and spiritual subject matter, trained on a synthetic dataset covering mysticism, hermeticism, and related topics. This model is designed for creative and exploratory work, offering an engaged, in-domain style for prompts within these specific areas.
Loading preview...
Model Overview
ewald1976/g4-12b-it-trismegistus is a 12 billion parameter instruction-tuned model, built upon the unsloth/gemma-4-12b-it base. It was fine-tuned using LoRA, which was then merged into the base weights, leveraging the Unsloth framework for efficient training.
Key Capabilities & Specialization
This model's primary distinction lies in its fine-tuning on the meseca/trismegistus-5k-v0.1 dataset. This dataset, derived from Teknium's Trismegistus Project, comprises synthetically generated instructions focused on a broad range of esoteric subjects, including:
- Mysticism and Hermeticism
- Religion and Spirituality
- Meditation and Magick
- Alchemy, Numerology, and Tarot
As a result, the model adopts an engaged and in-domain tone when responding to prompts within these areas, rather than a detached, encyclopedic style. While it largely retains the general instruction-following capabilities of its base model, its specialized training shifts its focus and tone towards these specific domains.
Intended Use Cases
This model is particularly well-suited for creative and exploratory applications involving esoteric, occult, and spiritual themes. Potential uses include:
- Worldbuilding: Generating lore and concepts for fictional settings.
- Thematic Writing: Crafting narratives or poetry with spiritual or mystical undertones.
- Conversational Exploration: Engaging in discussions about complex esoteric concepts.
Limitations
It is important to note that the training data is entirely synthetic and not factually authoritative. Therefore, the model's outputs should not be treated as factual reference material. Its specialized focus may also reduce its neutrality on esoteric topics.