jayfurzy/paterikon-sft-v2

TEXT GENERATIONConcurrent Unit Cost:1Model Size:3.1BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Jun 5, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

jayfurzy/paterikon-sft-v2 is a 3.09 billion parameter instruction-tuned causal language model built on Qwen2.5-3B CPT, developed by Justin Fursov. This model underwent three iterative rounds of supervised fine-tuning with active learning, resulting in 3,287 high-quality Q&A pairs. It is specifically optimized for Orthodox theological Q&A, spiritual guidance, and patristic text generation, primarily in English but also supporting Russian and Greek.

Loading preview...

Overview

Paterikon-SFT-v2 is an instruction-tuned language model developed by Justin Fursov, specifically designed for Orthodox Christian theological applications. Built upon the jayfurzy/paterikon-3b base model (a Qwen2.5-3B CPT variant), this 3.09 billion parameter model has undergone a unique three-pass iterative Supervised Fine-Tuning (SFT) process using active learning.

Key Capabilities

  • Theological Accuracy: Refined through an active learning loop where a judge evaluates model-generated responses for theological accuracy, pastoral tone, and attribution fidelity.
  • Iterative Training: Utilizes 3,287 high-quality Q&A pairs generated across three iterations, with each round building on the improved model from the previous.
  • Multilingual Support: Primarily English, with additional support for Russian and Greek.
  • Patristic Domain Fluency: Derived from a base model (paterikon-3b) already specialized in patristic texts.

Intended Use Cases

  • Orthodox theological Q&A with a pastoral voice.
  • Spiritual guidance and counseling.
  • Generation and completion of patristic texts.
  • Educational tool for Orthodox Christian studies.
  • Recall and attribution of Church Father quotes.

Limitations

While highly specialized, the model may still exhibit confabulation regarding specific historical dates or exact citations. It is not intended as a substitute for human spiritual guidance and reflects biases from its synthetic, primarily Russian Orthodox, training data. Due to its 3B parameter size, it has limitations compared to larger models for highly complex theological reasoning.

Finetunes
1 models