VERBAREX

We focus on fine-tuning large language models for improved factual consistency and instruction adherence.

Our primary goal is bridging the gap between raw base models and practical, stable assistants. We test and release models optimized for specific behavioral constraints, minimizing hallucinations in open-ended generation.

Current Projects

• LuminoLex Series: A set of fine-tuned models based on the Qwen architecture, optimized for strict identity retention and factual accuracy.
  • LuminoLex-14B: Our flagship instruct-tuned model.

Roadmap

• Foundation Models: We are currently establishing data pipelines and infrastructure to pre-train custom architectures from scratch.