Rocinante 12B is a 12-billion-parameter fine-tuned model for roleplay and adventure storytelling, created by TheDrummer and presented by BeaverAI. Its Hugging Face cards describe it as "a versatile workhorse for any adventure". There are two releases: Rocinante-12B-v1 and Rocinante-12B-v1.1. The v1 card recommends v1.1 for a more reliable experience, so this entry links to v1.1 as the main release.
Key features
- Story and roleplay focus: the cards point to roleplay and adventure storytelling as the intended use.
- Three chat templates: ChatML for roleplay, Alpaca for story or instruct-style adventures, and Mistral for NeMo. The cards suggest mixing them to see which works best.
- Sampler advice: use DRY. A temperature of 0.7 gives a calmer style, and 1.2 a more energetic one.
- v1.1 vs v1: the v1.1 card gives no technical changelog. It quotes tester feedback about richer, more distinct prose and cranked-up creativity, and recommends v1 for a "pure off-the-rails experience". This is subjective feedback, not a benchmark.
- Quantizations: GGUF builds are linked from the cards. v1.1 also lists EXL2 quants by Statuo at 4, 6 and 8 bits per weight.
Model specs
| Spec | Rocinante 12B |
|---|---|
| Parameters | 12B (BF16, Safetensors) |
| Releases | v1 and v1.1 |
| License | Listed as "other" on Hugging Face |
| Base model | Not stated on the cards |
| Hugging Face repos | TheDrummer/Rocinante-12B-v1, TheDrummer/Rocinante-12B-v1.1 |
The cards do not state a release date, a context length or a base model, so this entry does not either. They also note that no Inference Provider currently hosts the model, so you run it yourself.
Content note
The cards say testers reported "new and moistier words", which signals that the model can produce more explicit content. Use it in settings where that is appropriate, and add your own filtering for public products.
Use cases
- Interactive fiction and adventure games.
- Character roleplay in local chat front ends.
- Creative writing where a small, self-hostable model is enough.
Getting started
- Download the weights or a GGUF or EXL2 quantization from the v1.1 card.
- Pick a chat template (ChatML for roleplay, Alpaca for story mode) and enable DRY sampling.
- Start at temperature 0.7 for a calmer tone and raise it toward 1.2 for more energy.
FAQ
What is Rocinante 12B for?
Roleplay and adventure storytelling.
Which version should I use?
The v1 card recommends v1.1 for a more reliable experience, and recommends v1 for a pure off-the-rails experience.
What license does it use?
Hugging Face lists the license as "other". Read the card before commercial use.
Is it hosted by an API provider?
The cards say no Inference Provider currently deploys it.
Alternatives
- Mistral Nemo: the 12B Mistral model, whose template the cards reference.
- Gemma 4 26B A4B: a current open model for general chat.
- MythoMax 13B: another community model aimed at story and roleplay.
Conclusion
Rocinante 12B is a small, community-tuned model for fiction and roleplay, with clear template and sampler guidance on its cards. Start from the v1.1 card, and browse more models in the models category.
Sources
Comments
No comments yet. Be the first to comment!