đ Estopia Model
Estopia is a model designed to enhance dialogue and prose in instruct - format usage, with improvements in character cards and guided narratives.
đ Quick Start
The Estopia model aims to improve the quality of dialogue and prose when using the instruct format. It excels in "guided narratives" and has been tested in various modes like instruct, chat, adventure, and story modes.
⨠Features
- Enhanced Dialogue and Prose: Focuses on improving the dialogue and prose returned in instruct - format usage.
- Guided Narratives: Specializes in using instructions to guide or explore fictional stories.
- Multi - mode Compatibility: Capable of handling instruct, chat, adventure, and story modes to a certain degree, with strengths in instruct and adventure modes.
đĻ Installation
No installation steps are provided in the original document, so this section is skipped.
đģ Usage Examples
Basic Usage
The Estopia model has been tested with multiple formats. Here are some examples:
"\n### Instruction:\nWhat colour is the sky?\n### Response:\nThe sky is..."
"<Story text>\n***\nWrite a summary of the text above\n***\nThe story starts by..."
"User:Hello there!\nAssistant:Good morning..."
Advanced Usage
For general settings, the following are recommended:
{
"Temperature": "0.8 - 1.2",
"Min P": "0.05 - 0.1",
"Max P": "0.92, or 1 if using a Min P greater than 0",
"Top K": "0",
"Response length": "Higher than usual, e.g., 512",
"Rep Pen": "1.1",
"Rep Pen Range": "2 or 3x your response length",
"Stopping tokens": "##||$||---||$||ASSISTANT:||$||[End||$||</s>"
}
đ Documentation
Recipe
This model was created in three stages:
EstopiaV9
merge_method: task_arithmetic
base_model: TheBloke/Llama-2-13B-fp16
models:
- model: TheBloke/Llama-2-13B-fp16
- model: Undi95/UtopiaXL-13B
parameters:
weight: 1.0
- model: Doctor-Shotgun/cat-v1.0-13b
parameters:
weight: 0.02
- model: PygmalionAI/mythalion-13b
parameters:
weight: 0.10
- model: Undi95/Emerhyst-13B
parameters:
weight: 0.05
- model: CalderaAI/13B-Thorns-l2
parameters:
weight: 0.05
- model: KoboldAI/LLaMA2-13B-Tiefighter
parameters:
weight: 0.20
dtype: float16
EstopiaV13
merge_method: task_arithmetic
base_model: TheBloke/Llama-2-13B-fp16
models:
- model: TheBloke/Llama-2-13B-fp16
- model: Undi95/UtopiaXL-13B
parameters:
weight: 1.0
- model: Doctor-Shotgun/cat-v1.0-13b
parameters:
weight: 0.01
- model: chargoddard/rpguild-chatml-13b
parameters:
weight: 0.02
- model: PygmalionAI/mythalion-13b
parameters:
weight: 0.08
- model: CalderaAI/13B-Thorns-l2
parameters:
weight: 0.02
- model: KoboldAI/LLaMA2-13B-Tiefighter
parameters:
weight: 0.20
dtype: float16
Final Merge
models:
- model: EstopiaV9
parameters:
weight: 1
density: 1
- model: EstopiaV13
parameters:
weight: 0.05
density: 0.30
merge_method: dare_ties
base_model: TheBloke/Llama-2-13B-fp16
parameters:
int8_mask: true
dtype: bfloat16
Model Selection
Property |
Details |
Undi95/UtopiaXL - 13B |
A solid all - around base for models, capable of writing longer responses and retaining details well. |
Doctor - Shotgun/cat - v1.0 - 13b |
A medical - focused model that adds more human - like responses, e.g., for psychology. |
PygmalionAI/mythalion - 13b |
A role - play and instruct - focused model that improves attention to character card details and response variety. |
Undi95/Emerhyst - 13B |
A role - play model with longer - form responses, adding depth and more response options during narratives. |
CalderaAI/13B - Thorns - l2 |
A neutral and attentive model, good at chat and following instructions. |
KoboldAI/LLaMA2 - 13B - Tiefighter |
A well - rounded model focusing on story writing and adventure modes, enhancing creativity and prose. |
chargoddard/rpguild - chatml - 13b |
A role - play model that introduces new data and improves detail retention in long narratives. |
Notes
â ī¸ Important Note
Due to the different models inside, this model may not have perfect end - of - sequence tokens. You may encounter oddly behaving tokens, which can usually be resolved with a quick manual edit.
Chat is one of the least - tested areas for this model, and its performance can be quite character - card dependent.
This is a narrative and prose - focused model. It may talk extensively if guided to do so, which can be mitigated by adding instructions to limit it or using chat mode.
Future Areas
- Llava: Merging the llava lora has shown some success, but it also has drawbacks such as degraded performance in other areas and hallucinations.
- Stheno: Merge attempts between Estopia and Stheno are promising but not yet fully consistent.
- DynamicFactor: A dare merge with dynamic factor showed an improvement in instruction abilities and response length, but with a slight loss in response variety.
đ License
This model is licensed under cc - by - nc - 4.0.