farbodtavakkoli/OTel-LLM-7B-IT

OTel-LLM-7B-IT: A Specialized Telecom Language Model

OTel-LLM-7B-IT is a 7 billion parameter language model developed by farbodtavakkoli, built upon the allenai/OLMo-3-7B base model. It is a full-parameter fine-tuned model, specifically designed for the telecommunications sector as part of the Open Telco AI initiative.

Key Capabilities & Differentiators

• Domain-Specific Expertise: Fine-tuned on the proprietary OTel-LLM dataset, curated by over 100 domain experts, including data from arXiv telecom papers, 3GPP standards, GSMA documents, and O-RAN specifications.
• Context-Grounded Accuracy: Demonstrates significant improvement in context-grounded correctness, with a +6.0 percentage point delta over its base model (57.4% to 63.4% +/- 0.8) on held-out OTel evaluation data, as judged by GPT-4o mini.
• RAG Optimization: Primarily intended for Retrieval-Augmented Generation (RAG) pipelines, where it generates answers grounded in provided telecom context.
• Robust Training: Utilizes a comprehensive training recipe including AdamW 8-bit optimizer, Flash Attention 2, and BF16 precision, trained on AMD MI300X/MI325X/MI355X and NVIDIA A100/H100 GPUs.

Intended Use Cases

• Telecom Answer Generation: Ideal for applications requiring precise, context-grounded answers to telecommunications-related queries.
• RAG Systems: Best suited for integration into RAG architectures where external telecom context is provided to the model.

Limitations

• Domain-Specific: Not a general-purpose LLM; performance outside the telecom domain is not guaranteed.
• English-Only: The current release is text-centric and supports only the English language.
• Context-Dependent: Not optimized for unrestricted, context-free question answering. Users should implement abstention mechanisms for queries lacking sufficient retrieved context.