farbodtavakkoli/OTel-LLM-8.3B-Safety

OTel-LLM-8.3B-Safety: A Safety-Tuned Telecom Language Model

OTel-LLM-8.3B-Safety is an 8.3 billion parameter language model developed by Farbod Tavakkoli, specifically fine-tuned for safety within telecommunications applications. It is part of the broader OTel Family of Models, an open-source initiative focused on creating AI resources for the global telecom sector. This model is built upon the EssentialAI/rnj-1-instruct base model and underwent full-parameter post-training using the specialized OTel-Safety dataset.

Key Capabilities

• Contextual Abstention: Designed to refuse to answer when the retrieved context is insufficient, irrelevant, or off-topic, improving the reliability of responses in RAG systems.
• Telecom Domain Expertise: Fine-tuned on a comprehensive dataset curated by over 100 domain experts, covering arXiv telecom papers, 3GPP standards, GSMA documents, IETF RFCs, and O-RAN specifications.
• Auxiliary Safety Model: Functions as an auxiliary model to enhance safety and precision in telecom RAG pipelines, focusing on abstention behavior rather than general-purpose generation.

Good For

• Telecom RAG Pipelines: Ideal for deployment in Retrieval Augmented Generation (RAG) systems within the telecommunications industry where accurate, context-bound answers are critical.
• Ensuring Response Reliability: Useful for applications requiring the model to explicitly state when it lacks sufficient information to provide a confident answer.
• Reference for Further Safety Tuning: Can serve as a checkpoint for additional safety-focused fine-tuning tailored to specific deployment scenarios.

It is important to note that this model is domain-specific to telecommunications, English-only, and its abstention quality depends heavily on the surrounding retrieval, reranking, context window, and prompt policies.