155 lines
3.3 KiB
Markdown
155 lines
3.3 KiB
Markdown
---
|
|
license: apache-2.0
|
|
|
|
base_model: Qwen/Qwen2.5-7B-Instruct
|
|
|
|
library_name: llama.cpp
|
|
|
|
pipeline_tag: text-generation
|
|
|
|
language:
|
|
- de
|
|
- en
|
|
|
|
tags:
|
|
- gguf
|
|
- qwen2.5
|
|
- history
|
|
- archaeology
|
|
- ancient-texts
|
|
- religious-studies
|
|
- biblical-studies
|
|
- research-assistant
|
|
- local-ai
|
|
- german
|
|
- english
|
|
- lm-studio
|
|
---
|
|
|
|
# Chaperon-Sophia-v2
|
|
|
|
An experimental research assistant for archaeology, history, biblical studies, and ancient texts.
|
|
|
|
Developed by **Chaperon GmbH** and **Lucas Bischof**, Sophia-v2 demonstrates a complete end-to-end AI workflow including multimodal data processing, document extraction, dataset generation, LoRA fine-tuning, GGUF deployment, and local AI integration.
|
|
|
|
## Overview
|
|
|
|
Sophia-v2 is a custom fine-tuned variant of **Qwen2.5-7B-Instruct** designed to explore historical, textual, archaeological, and religious source material through natural conversation.
|
|
|
|
The project serves both as a technical showcase of Chaperon GmbH's AI capabilities and as a practical proof-of-concept for domain-specific model adaptation.
|
|
|
|
## Project Goals
|
|
|
|
This project demonstrates the ability to:
|
|
|
|
- Process large collections of PDFs
|
|
- Extract text from historical documents
|
|
- Convert audio and video into training datasets
|
|
- Process image-based source material
|
|
- Generate structured instruction datasets
|
|
- Fine-tune Large Language Models using LoRA
|
|
- Merge and deploy custom models
|
|
- Publish GGUF models for local inference
|
|
|
|
## Training Sources
|
|
|
|
The training process included the preparation and processing of curated material related to:
|
|
|
|
- CDLI Cuneiform Collections
|
|
- ETCSL Translations
|
|
- Ancient Mesopotamian Literature
|
|
- Biblical Texts
|
|
- Apocryphal Literature
|
|
- Gnostic Literature
|
|
- Gospel of Mary
|
|
- Pistis Sophia
|
|
- Comparative Religion
|
|
- Historical Source Material
|
|
- Archaeological Research
|
|
|
|
The focus of Sophia-v2 is source-oriented exploration and discussion rather than authoritative interpretation.
|
|
|
|
## Intended Audience
|
|
|
|
Sophia-v2 may be useful for:
|
|
|
|
- Archaeology enthusiasts
|
|
- History researchers
|
|
- Biblical scholars
|
|
- Comparative religion communities
|
|
- Independent researchers
|
|
- Educational projects
|
|
- Digital humanities initiatives
|
|
|
|
## Available Models
|
|
|
|
### Sophia_v2_q5_K_M.gguf
|
|
|
|
Recommended version for most users.
|
|
|
|
Provides an excellent balance between quality, performance, and memory usage.
|
|
|
|
### Sophia_v2_f16.gguf
|
|
|
|
Reference-quality version.
|
|
|
|
Provides maximum fidelity while requiring significantly more storage and memory.
|
|
|
|
## Technical Specifications
|
|
|
|
Base Model:
|
|
|
|
- Qwen/Qwen2.5-7B-Instruct
|
|
|
|
Training Method:
|
|
|
|
- Supervised Fine Tuning (SFT)
|
|
- LoRA Adaptation
|
|
- Model Merge
|
|
- GGUF Conversion
|
|
|
|
Deployment Targets:
|
|
|
|
- LM Studio
|
|
- llama.cpp
|
|
- Open WebUI
|
|
- KoboldCpp
|
|
|
|
## Limitations
|
|
|
|
Sophia-v2 is a language model and may:
|
|
|
|
- Generate incorrect information
|
|
- Misinterpret source material
|
|
- Produce inaccurate citations
|
|
- Present plausible but incorrect conclusions
|
|
|
|
Users should verify important claims using primary and scholarly sources.
|
|
|
|
## About Chaperon GmbH
|
|
|
|
Sophia-v2 serves as a public demonstration of the capabilities of Chaperon GmbH in:
|
|
|
|
- Multimodal data processing
|
|
- Knowledge extraction
|
|
- AI training pipelines
|
|
- Custom model development
|
|
- Local AI deployment
|
|
- Enterprise AI integration
|
|
|
|
## Credits
|
|
|
|
Created by:
|
|
|
|
**Lucas Bischof**
|
|
**Chaperon GmbH**
|
|
|
|
Base Model:
|
|
|
|
**Qwen/Qwen2.5-7B-Instruct**
|
|
|
|
GGUF Conversion:
|
|
|
|
**llama.cpp**
|
|
>>>>>>> 038febb (Improve model card)
|