plvictor/Alef-Biblical-1.5B-Base

TEXT GENERATIONConcurrency Cost:1Model Size:2BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Jun 6, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

The Alef Biblical 1.5B Base model, developed by Paulo Victor Souza, is a 1.51 billion parameter Qwen3-based causal language model specialized in Brazilian Portuguese for biblical, theological, and Christian faith content. This base model, with a 4096-token context length, underwent continued pre-training on a curated ~368 million token Portuguese biblical corpus. It is designed for text completion tasks within its specific domain, serving as the foundation for the conversational Alef-Biblical-1.5B-Instruct.

Loading preview...

Alef Biblical 1.5B Base: Specialized Portuguese Biblical AI

Alef Biblical 1.5B Base is a 1.51 billion parameter Qwen3-based language model developed by Paulo Victor Souza, specifically tailored for the Brazilian Portuguese-speaking community. Unlike most specialized biblical AI models that are English-centric, Alef aims to fill a niche for Portuguese content in theology and Christian faith.

Key Characteristics & Training

  • Architecture: Based on the Qwen3 family, specifically using Polygl0t/Tucano2-qwen-1.5B-Base as its foundation.
  • Language & Domain: Exclusively focused on Brazilian Portuguese (PT-BR) within the domain of the Bible, theology, and Christian faith.
  • Continued Pre-Training (CPT): The model underwent CPT on a curated dataset of approximately 368 million tokens. This corpus includes a diverse range of biblical content such as devotionals, reading plans, biblical studies, Q&A, multiple Bible translations, and content from theological websites like voltemosaoevangelho.com and bibliotecadopregador.com.br.
  • Context Length: Features a context window of 4096 tokens.
  • Developer's Vision: Developed by a solo Brazilian developer, this project emphasizes demonstrating the feasibility of specialized fine-tuning with limited resources and fostering the growth of the Brazilian AI community.

Use Cases & Limitations

  • Primary Use: This is a base model intended for text completion tasks within its specialized domain. For conversational or assistant-like interactions, the instruction-tuned variant, plvictor/Alef-Biblical-1.5B-Instruct, is recommended.
  • Limitations: Due to its 1.5B parameter size, it is best suited for specific biblical queries and conversations rather than complex reasoning tasks. It is not designed for general-purpose use and may reflect theological perspectives prevalent in its training data.