70 lines
3.4 KiB
Markdown
70 lines
3.4 KiB
Markdown
---
|
|
license: apache-2.0
|
|
language:
|
|
- en
|
|
base_model:
|
|
- mistralai/Mistral-Nemo-Base-2407
|
|
tags:
|
|
- text adventure
|
|
- roleplay
|
|
library_name: transformers
|
|
---
|
|
|
|
|
|
|
|
# Muse-12B
|
|
|
|
Muse brings an extra dimension to any tale—whether you're exploring a fantastical realm, court intrigue, or slice-of-life scenarios where a conversation can be as meaningful as a quest. While it handles adventure capably, Muse truly shines when character relationships and emotions are at the forefront, delivering impressive narrative coherence over long contexts.
|
|
|
|
If you want to easily try this model for free, you can do so at [https://aidungeon.com](https://aidungeon.com/).
|
|
|
|
We plan to continue improving and open-sourcing similar models, so please share any and all feedback on how we can improve model behavior. Below we share more details on how Muse was created.
|
|
|
|
[Quantized GGUF weights can be downloaded here.](https://huggingface.co/LatitudeGames/Muse-12B-GGUF)
|
|
|
|
## Model details
|
|
|
|
Muse 12B was trained using Mistral Nemo 12B as its foundation, with training occurring in three stages: SFT (supervised fine-tuning), followed by two distinct DPO (direct preference optimization) phases.
|
|
|
|
**SFT** - Various multi-turn datasets from a multitude of sources, combining text adventures of the kind used to finetune [our Wayfarer 12B model](https://huggingface.co/LatitudeGames/Wayfarer-12B), long emotional narratives and general roleplay, each carefully balanced and rewritten to be free of common AI cliches. A small single-turn instruct dataset was included to send a stronger signal during finetuning.
|
|
|
|
**DPO 1** - Gutenberg DPO, [credit to Jon Durbin](https://huggingface.co/datasets/jondurbin/gutenberg-dpo-v0.1) - This stage introduces human writing techniques, significantly enhancing the model's potential outputs, albeit trading some intelligence for the stylistic benefits of human-created text.
|
|
|
|
**DPO 2** - Reward Model User Preference Data, [detailed in our blog](https://blog.latitude.io/all-posts/synthetic-data-preference-optimization-and-reward-models) - This stage refines the Gutenberg stage's "wildness," restoring intelligence while maintaining enhanced writing quality and providing a final level of enhancement due to the reward model samples.
|
|
|
|
The result is a model that writes like no other: versatile across genres, natural in expression, and suited to emotional depth.
|
|
|
|
## Inference
|
|
|
|
The Nemo architecture is known for being sensitive to higher temperatures, so the following settings are recommended as a baseline. Nothing stops you from experimenting with these, of course.
|
|
|
|
```
|
|
"temperature": 0.8,
|
|
"repetition_penalty": 1.05,
|
|
"min_p": 0.025
|
|
```
|
|
|
|
## Limitations
|
|
|
|
Muse was trained exclusively on second-person present tense data (using “you”) in a narrative style. Other styles will work as well but may produce suboptimal results.
|
|
|
|
Average response lengths tend toward verbosity (1000+ tokens) due to the Gutenberg DPO influence, though this can be controlled through explicit instructions in the system prompt.
|
|
|
|
## Prompt Format
|
|
|
|
ChatML was used during all training stages.
|
|
|
|
```
|
|
<|im_start|>system
|
|
You're a masterful storyteller and gamemaster. Write in second person present tense (You are), crafting vivid, engaging narratives with authority and confidence.<|im_end|>
|
|
<|im_start|>user
|
|
> You peer into the darkness.
|
|
<|im_start|>assistant
|
|
You have been eaten by a grue.
|
|
|
|
GAME OVER
|
|
```
|
|
|
|
## Credits
|
|
|
|
Thanks to [Gryphe Padar](https://huggingface.co/Gryphe) for collaborating on this finetune with us! |