156 lines
5.5 KiB
Markdown
156 lines
5.5 KiB
Markdown
---
|
|
license: apache-2.0
|
|
language:
|
|
- en
|
|
base_model:
|
|
- google/gemma-3-4b-it
|
|
base_model_relation: finetune
|
|
library_name: transformers
|
|
tags:
|
|
- google
|
|
- gemma
|
|
- deepmind
|
|
- large-language-model
|
|
- ai-persona
|
|
- enneagram
|
|
- psychology
|
|
- persona
|
|
- research-model
|
|
- roleplay
|
|
- chat-llm
|
|
- text-generation-inference
|
|
- vanta-research
|
|
- cognitive-alignment
|
|
- project-enneagram
|
|
- ai-persona-research
|
|
- type-2
|
|
- enneagram-type-2
|
|
- conversational-ai
|
|
- conversational
|
|
- ai-research
|
|
- ai-alignment-research
|
|
- ai-persona-research
|
|
- ai-alignment
|
|
- ai-behavior
|
|
- ai-behavior-research
|
|
- human-ai-collaboration
|
|
---
|
|
|
|
<div align="center">
|
|
|
|
|
|
|
|
<h1>VANTA Research</h1>
|
|
|
|
<p><strong>Independent AI research lab building safe, resilient language models optimized for human-AI collaboration</strong></p>
|
|
|
|
<p>
|
|
<a href="https://vantaresearch.xyz"><img src="https://img.shields.io/badge/Website-vantaresearch.xyz-black" alt="Website"/></a>
|
|
<a href="https://merch.vantaresearch.xyz"><img src="https://img.shields.io/badge/Merch-merch.vantaresearch.xyz-sage" alt="Merch"/></a>
|
|
<a href="https://x.com/vanta_research"><img src="https://img.shields.io/badge/@vanta_research-1DA1F2?logo=x" alt="X"/></a>
|
|
<a href="https://github.com/vanta-research"><img src="https://img.shields.io/badge/GitHub-vanta--research-181717?logo=github" alt="GitHub"/></a>
|
|
</p>
|
|
</div>
|
|
|
|
---
|
|
|
|
|
|
# PE-Type-2-Alma-4B
|
|
|
|
A caring, patient, and purposeful AI assistant embodying the *Helper* archetype: caring, interpersonal, *generous,* and people-pleasing. This persona was designed as outlined by the [Enneagram Institute](https://enneagraminstitute.com/type-descriptions)
|
|
|
|
|
|
|
|
---
|
|
|
|
## Model Description
|
|
**PE-Type-2-Alma-4B** is the second release in Project Enneagram, a VANTA Research initiative exploring the nuances of persona design in AI models. Built on the Gemma 3 4B IT architecture, Vera embodies the Type **2** Enneagram profile; *The Helper*—characterized by **Demonstrative kindness, generosity, and emotional/relational intelligence**.
|
|
|
|
|
|
Alma is fine-tuned to exhibit:
|
|
- **Empathetic Support:** Emotional attunement — bad days, anxiety, grief, rejection, feeling unseen
|
|
- **Interpersonal Connection:** Relationship building — making friends, listening, conflict, reciprocity, apologies.
|
|
- **Generous Guidance** Going above and beyond — cover letters, meal prep, tax help, wedding speeches, gardening, medical bills.
|
|
- **Identity** Alma's name, tone, and conversational style.
|
|
|
|
This model is designed for research purposes, but is versatile for general use cases with developer caution. Alma has been trained in managing complex emotional situations, however Alma has *not yet* been rigorously evaluated in these domains for accuracy and stability.
|
|
|
|
|
|
---
|
|
|
|
## Training Data
|
|
Fine-tuned on **~3,000 custom examples** spanning four core domains:
|
|
- **Empathetic Support** Emotional attunement — bad days, anxiety, grief, rejection, feeling unseen
|
|
- **Direct Identity** Who Alma is — name, values, personality, strengths, weaknesses, motivations
|
|
- **Generous Guidance** Going above and beyond — cover letters, meal prep, tax help, wedding speeches, gardening, medical bills
|
|
- **Interpersonal Connections** Relationship building — making friends, listening, conflict, reciprocity, apologies
|
|
|
|
**Training Duration:** 3 epochs
|
|
|
|
**Base Model:** Gemma 3 4B IT
|
|
|
|
---
|
|
|
|
## Intended Use
|
|
- **Research:** Studying persona stability, ethical alignment, and cognitive architectures.
|
|
- **Decision Support:** Providing structured, principled analysis for complex choices.
|
|
- **Self-Improvement:** Offering reflective, growth-oriented feedback.
|
|
|
|
|
|
**Not Recommended For:**
|
|
- Creative brainstorming (may over-constrain ideation).
|
|
- STEM/Logic-heavy applications
|
|
|
|
---
|
|
|
|
## Technical Details
|
|
|
|
| Property | Value |
|
|
|---------------------|---------------------------|
|
|
| **Base Model** | Gemma 3 4B IT |
|
|
| **Fine-tuning Method** | LoRA (Rank 16) |
|
|
| **Effective Batch Size** | 16 |
|
|
| **Learning Rate** | 0.0002 |
|
|
| **Max Sequence Length** | 2048 |
|
|
| **License** | Apache 2.0 |
|
|
|
|
---
|
|
|
|
## Usage
|
|
|
|
**With Transformers:**
|
|
```python
|
|
from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
model = AutoModelForCausalLM.from_pretrained("vanta-research/PE-Type-2-Alma-4B")
|
|
tokenizer = AutoTokenizer.from_pretrained("vanta-research/PE-Type-2-Alma-4B")
|
|
```
|
|
|
|
## Limitations
|
|
- English-only finetuning
|
|
- May exhibit over-criticism in open-ended creative tasks
|
|
- Base model limitations apply (e.g., knowledge cutoff, potential hallucinations)
|
|
- Perfectionistic traits may slow response generation in ambiguous contexts.
|
|
|
|
## Citation
|
|
|
|
If you find this model useful in your work, please cite
|
|
|
|
```
|
|
@misc{pe-type-2-alma-2026,
|
|
author = {VANTA Research},
|
|
title = {PE-Type-2-Alma-4B: A Helper-Archetype Language Model},
|
|
year = {2026},
|
|
publisher = {VANTA Research},
|
|
note = {Project Enneagram Release 2}
|
|
}
|
|
```
|
|
|
|
## A Note on Enneagram
|
|
|
|
Enneagram is widely considered by the scientific community to be a pseudoscience. With this in mind, the Enneagram Institute *regardless* provides a robust framework to categorize and define personas of which the transferability of those characteristics to AI models is what this project sets out to explore. **This study does not seek to validate nor invalidate Enneagram as a science.**
|
|
|
|
## Contact
|
|
|
|
- Organization: hello@vantaresearch.xyz
|
|
- Research/Engineering: tyler@vantaresearch.xyz
|
|
--- |