farbodtavakkoli/OTel-LLM-12B-Safety
OTel-LLM-12B-Safety by farbodtavakkoli is a 12 billion parameter language model fine-tuned on telecommunications domain data, based on google/gemma-3-12b-it. It is specifically designed for telecommunications applications, excelling at generating accurate responses grounded in retrieved context. This model is part of the OTel Family, an open-source initiative for industry-standard AI in the global telecom sector, and features abstention training to prevent hallucination when context is insufficient.
Loading preview...
What is OTel-LLM-12B-Safety?
OTel-LLM-12B-Safety is a 12 billion parameter language model developed by farbodtavakkoli, built upon the google/gemma-3-12b-it base model. It is a specialized model within the OTel Family, an open-source effort to create AI solutions for the telecommunications industry. The model underwent full parameter fine-tuning using a comprehensive dataset curated by over 100 domain experts from institutions like Yale University, GSMA, NetoAI, Khalifa University, University of Leeds, and The University of Texas at Dallas, covering sources such as arXiv telecom papers, 3GPP standards, GSMA documents, IETF RFCs, and O-RAN specifications.
Key Capabilities
- Telecommunications Specialization: Fine-tuned on extensive telecom-specific data for high relevance in the domain.
- Context-Grounded Generation: Optimized to generate accurate responses based on provided context, making it suitable for Retrieval-Augmented Generation (RAG) pipelines.
- Abstention Training: Designed to decline answering when insufficient context is provided, actively mitigating hallucination.
- Apache 2.0 License: Available for broad use under an open-source license.
Intended Use Cases
This model is primarily intended to power end-to-end RAG pipelines for telecommunications, either as a standalone LLM or integrated with other OTel models (Embedding and Reranker). It is particularly well-suited for applications requiring precise, fact-based answers within the telecom domain, where preventing hallucination is critical. Users should leverage its context-grounded nature rather than expecting open-ended conversational abilities.