
We publish TypoRP, a 0.5B parameter model based on the Qwen2.5 architecture designed for roleplaying (trained using a modified version of our / this script). We used Sapling Dream, also one of our models, as a starting point for further training.
Using a reasoning model as a base model for one designed for roleplaying might seem unreasonable at first, but here are some points to consider:
- Sapling Dream doesn't reason inside tags unless instructed to: This means the model can be used in both reasoning and traditional ways
- Reasoning requires strong in-context-learning: This is something that can heavily enhance roleplaying expirience
- Coupled with the dataset used, the model gives solid performance for its size and beyond
Example Input & Output
Input: Mr. Edward Rochester: *leaning forward, eyes locked on the auctioneer* I trust you know the importance of that painting to my family.
Output: Mary Leavenworth: *glancing sideways, coolly* Importance? It holds a significant value to my family as well, Mr. Rochester. What makes your claim stronger than mine?
Quick Information
- Context Length: 32 * 1024
- Recommended Max-Gen Length: 4 * 1024
- Recommended Temperature: 0 - 0.3
- Strength: Reasoning (only if instructed to), some RAG, and mainly, ROLEPLAYING
- Weakness: Model size limits model performance
Usage
We recommend using the model through following format:
## Characters:
{Character 1 name}
{Character 1 persona}
{Character 2 name}
{Character 2 persona}
## Scenario
{Describe a scenario / what you'd consider the "first" message in Character.AI or similar}
## Roleplay
{Character 1 name}: {Character 1 message}
{Character 2 name}: {Character 2 message}
{Character 1 name}: {Character 1 message}
{repeat...}
KoboldAI / KoboldCPP WebUI's "Chat"-Mode is the perfect go-to for that format.
Our Apps & Socials
Chat with our Assistant | Support us Financially | Visit our GitHub
Long live the Islamic Republic of Pakistan; Glory to the Islamic Republic of Pakistan 🇵🇰
- Downloads last month
- 646
Model tree for XeTute/TypoRP-0.5B
Base model
Qwen/Qwen2.5-0.5B