|
|
--- |
|
|
license: apache-2.0 |
|
|
tags: |
|
|
- merge |
|
|
- model_stock |
|
|
- Llama3 |
|
|
- Pneuma |
|
|
- Allades |
|
|
- storytelling |
|
|
- roleplaying |
|
|
- instruction-following |
|
|
- creative-writing |
|
|
- reasoning |
|
|
- long-form-generation |
|
|
- text-generation |
|
|
- lazymergekit |
|
|
- nbeerbower/Llama3.1-Allades-8B |
|
|
- mergekit-community/L3.1-Pneuma-8B-v1 |
|
|
base_model: |
|
|
- mergekit-community/L3.1-Pneuma-8B-v1 |
|
|
- nbeerbower/Llama3.1-Allades-8B |
|
|
pipeline_tag: text-generation |
|
|
--- |
|
|
|
|
|
|
|
|
# **L3.1-Pneuma-Allades-8B** |
|
|
|
|
|
**L3.1-Pneuma-Allades-8B** is a unique 8B parameter model, carefully merged using the **della_linear** method to integrate the best qualities of two exceptional models: **L3.1-Pneuma-8B-v1** and **Llama3.1-Allades-8B**. This model is finely tuned to handle advanced tasks in **creative writing**, **reasoning**, **roleplay**, and **instruction-following**. |
|
|
|
|
|
This model is designed for anyone seeking to enhance AI-assisted storytelling, detailed reasoning, or dynamic interactive dialogues. The combination of the Pneuma and Allades models ensures a well-balanced performance in both technical and creative realms. 🧠✨ |
|
|
|
|
|
--- |
|
|
|
|
|
## 🌐 **Family Tree** |
|
|
|
|
|
This model is the result of a fusion between the following: |
|
|
|
|
|
- [**mergekit-community/L3.1-Pneuma-8B-v1**](https://huggingface.co/mergekit-community/L3.1-Pneuma-8B-v1): A comprehensive 8B model designed to excel in logical reasoning and long-form content generation. It was merged with a focus on **instruction-following** and **roleplaying**. |
|
|
- [**nbeerbower/Llama3.1-Allades-8B**](https://huggingface.co/nbeerbower/Llama3.1-Allades-8B): A finetuned Llama 3.1 model, focused on **creative writing**, **reasoning**, and **roleplay**. Trained with a mix of public and private datasets to push the boundaries of creativity and interaction in AI language models. |
|
|
|
|
|
--- |
|
|
|
|
|
## 🧬 **Model Lineage** |
|
|
|
|
|
### **A: mergekit-community/L3.1-Pneuma-8B-v1** |
|
|
|
|
|
- Designed for advanced **reasoning** tasks, particularly in creative problem solving and interactive roleplay scenarios. |
|
|
- This model leverages the strength of **instruction-following** tasks with a highly optimized configuration, making it a robust choice for interactive experiences. |
|
|
|
|
|
### **B: nbeerbower/Llama3.1-Allades-8B** |
|
|
|
|
|
- Tuned with **creative writing**, **storytelling**, and **dynamic roleplay** in mind. |
|
|
- It integrates five datasets that focus on reasoning, storytelling, and narrative structure, providing a well-rounded experience for **creative and logical task performance**. |
|
|
|
|
|
--- |
|
|
|
|
|
## ⚙️ **Merge Details** |
|
|
|
|
|
The **L3.1-Pneuma-Allades-8B** model was merged using the **della_linear** method, ensuring a balanced integration of the two base models' distinct strengths. Here’s the exact configuration used for the merge: |
|
|
|
|
|
```yaml |
|
|
out_dtype: bfloat16 |
|
|
dtype: float32 |
|
|
tokenizer_source: base |
|
|
merge_method: della_linear |
|
|
parameters: |
|
|
int8_mask: true |
|
|
density: 0.5 |
|
|
epsilon: 0.04 |
|
|
lambda: 1.05 |
|
|
base_model: nbeerbower/Llama3.1-Allades-8B |
|
|
models: |
|
|
- model: nbeerbower/Llama3.1-Allades-8B |
|
|
parameters: |
|
|
weight: |
|
|
- filter: v_proj |
|
|
value: [1, 1, 0, 0, 0, 0, 0, 0, 0, 1, 1] |
|
|
- filter: o_proj |
|
|
value: [1, 1, 0, 0, 0, 0, 0, 0, 0, 1, 1] |
|
|
- filter: up_proj |
|
|
value: [1, 1, 0, 0, 0, 0, 0, 0, 0, 1, 1] |
|
|
- filter: gate_proj |
|
|
value: [1, 1, 0, 0, 0, 0, 0, 0, 0, 1, 1] |
|
|
- filter: down_proj |
|
|
value: [1, 1, 0, 0, 0, 0, 0, 0, 0, 1, 1] |
|
|
- value: 1 |
|
|
- model: mergekit-community/L3.1-Pneuma-8B-v1 |
|
|
parameters: |
|
|
weight: |
|
|
- filter: v_proj |
|
|
value: [0, 0, 1, 1, 1, 1, 1, 1, 1, 0, 0] |
|
|
- filter: o_proj |
|
|
value: [0, 0, 1, 1, 1, 1, 1, 1, 1, 0, 0] |
|
|
- filter: up_proj |
|
|
value: [0, 0, 1, 1, 1, 1, 1, 1, 1, 0, 0] |
|
|
- filter: gate_proj |
|
|
value: [0, 0, 1, 1, 1, 1, 1, 1, 1, 0, 0] |
|
|
- filter: down_proj |
|
|
value: [0, 0, 1, 1, 1, 1, 1, 1, 1, 0, 0] |
|
|
- value: 0 |
|
|
|
|
|
``` |
|
|
|
|
|
--- |
|
|
|
|
|
## 🔮 **Key Features & Capabilities** |
|
|
|
|
|
### **1. Instruction Following**: |
|
|
|
|
|
- **Pneuma's** structure allows this model to excel in **precise instruction-following** tasks, ensuring clear and coherent responses. |
|
|
|
|
|
### **2. Creative Writing & Storytelling**: |
|
|
|
|
|
- With **Allades'** fine-tuned datasets, this model is capable of generating detailed and creative narratives, perfect for roleplaying, interactive storytelling, and fantasy creation. |
|
|
|
|
|
### **3. Dynamic Roleplay**: |
|
|
|
|
|
- Combining the **reasoning power** of Pneuma with **Allades' focus** on roleplay and creativity, this model can generate interactive, logical, and emotionally-driven dialogues in any roleplay scenario. |
|
|
|
|
|
--- |
|
|
|
|
|
## 🚀 **Use Cases** |
|
|
|
|
|
- **Immersive Storytelling**: Whether you’re writing a novel, developing interactive character dialogue, or creating a magical world, this model offers both creativity and structure. |
|
|
- **Instructional Content**: From educational guides to complex instructional scenarios, this model adapts well to delivering step-by-step processes in a clear and structured format. |
|
|
- **AI-Assisted Roleplay**: Perfect for game development, NPC dialogues, and other roleplaying applications where both **logic** and **creativity** are required. |
|
|
|
|
|
--- |
|
|
|
|
|
## 📜 **License** |
|
|
|
|
|
This model is open-sourced under the **Apache-2.0 License**, allowing free use, modification, and distribution with proper attribution. |
|
|
|
|
|
--- |
