farbodtavakkoli/OTel-LLM-20B-IT

Hugging Face
TEXT GENERATIONConcurrent Unit Cost:2Model Size:20BQuant:FP8Context Size:128kPublished:Feb 13, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Warm

OTel-LLM-20B-IT is a 20 billion parameter instruction-tuned language model developed by farbodtavakkoli, based on the openai/gpt-oss-20b architecture. It is specifically fine-tuned on telecommunications data to improve context-grounded correctness in telecom-related tasks. This model is designed for use in Retrieval-Augmented Generation (RAG) pipelines for precise telecom answer generation.

Loading preview...

OTel-LLM-20B-IT: A Specialized Telecom Language Model

OTel-LLM-20B-IT is a 20 billion parameter language model, developed by farbodtavakkoli, that has been full-parameter fine-tuned on a curated dataset of telecommunications data. Built upon the openai/gpt-oss-20b base model, it is part of the OTel Family of Models, an open-source initiative focused on AI resources for the global telecommunications sector. The model demonstrates improved context-grounded correctness, showing a +5.0 percentage point increase over its base model in LLM-as-judge correctness on held-out OTel evaluation data.

Key Capabilities

  • Context-Grounded Telecom Answer Generation: Specifically optimized for generating answers based on provided telecom context within RAG pipelines.
  • Specialized Training Data: Trained on a dataset of 326,767 high-confidence telecom examples, curated by over 100 domain experts from various sources including arXiv papers, 3GPP standards, GSMA documents, and O-RAN specifications.
  • Full-Parameter Fine-Tuning: Utilizes full-parameter post-training on the OTel-LLM dataset to enhance domain-specific performance.

Good for

  • Retrieval-Augmented Generation (RAG) in Telecom: Ideal for applications requiring accurate, context-grounded responses to telecom-specific queries.
  • Domain-Specific QA Systems: Suitable for building question-answering systems where the input context is primarily telecommunications-related.

Limitations

  • Domain-Specific: Not intended as a general-purpose language model; performance is optimized for telecommunications.
  • English-Only: The current release supports only the English language.
  • Context-Dependent: Designed for context-grounded generation; performance for unrestricted, context-free telecom QA may vary and requires separate evaluation.