farbodtavakkoli/OTel-LLM-4B-IT

OTel-LLM-4B-IT: A Specialized Telecom Language Model

OTel-LLM-4B-IT is a 4 billion parameter instruction-tuned model developed by farbodtavakkoli, building upon Google's gemma-3-4b-it base. It is a key component of the OTel Family of Models, an open-source initiative focused on AI resources for the global telecommunications sector.

Key Capabilities & Differentiators

• Domain-Specific Optimization: Full-parameter fine-tuned on the OTel-LLM dataset, which comprises telecom-focused data curated by over 100 domain experts.
• Enhanced Context-Grounded Correctness: Achieves a +7.0 percentage point improvement in LLM-as-judge correctness (73.2% vs. 66.2%) over its base model for context-grounded telecom questions.
• RAG Pipeline Focus: Primarily intended for context-grounded telecom answer generation within Retrieval-Augmented Generation (RAG) pipelines, requiring retrieved context for optimal performance.

Training & Data

• Trained using full-parameter post-training on a filtered dataset of 326,767 high-confidence telecom examples.
• Training data sources include arXiv telecom papers, 3GPP standards, GSMA documents, IETF RFCs, and O-RAN specifications.

Intended Use Cases

• Telecom RAG Systems: Ideal for applications where the model receives retrieved telecom context and needs to generate an answer grounded in that context.
• Specialized Telecom QA: Suitable for precise, context-aware question answering within the telecommunications domain.

Limitations

• Domain-Specific: Not designed as a general-purpose language model; performance is optimized for telecommunications.
• Context-Dependent: Not optimized for unrestricted, context-free telecom QA. Users should provide retrieved context or use abstention-aware prompts for questions where context is missing.