farbodtavakkoli/OTel-2.0-LLM-31B-IT
The OTel-2.0-LLM-31B-IT is a 31 billion parameter instruction-tuned language model developed by AT&T Chief Data Office, post-trained from Gemma 4 31B-IT on approximately 440 billion telecom training tokens. This model is specialized for telecommunications, excelling in tasks like standards interpretation, network configuration assistance, and telecom-specific question answering. It leverages Orthogonal Subspace Fine-Tuning (OSFT) to absorb new domain knowledge while preserving general instruction-following capabilities, making it ideal for telecom-focused AI workflows.
Loading preview...
OTel-2.0-LLM-31B-IT: A Telecom-Specialized LLM
OTel-2.0-LLM-31B-IT is a 31 billion parameter instruction model, post-trained by AT&T Chief Data Office from Google's Gemma 4 31B-IT. It was trained on an extensive corpus of approximately 440 billion telecom-specific tokens, significantly expanding upon the original OTel effort. This model is the first release in the OTel 2.0 family, designed to support a wide range of telco-grade AI applications.
Key Capabilities
- Telecom Domain Adaptation: Specialized for telecommunications, trained on ~440 billion tokens from sources like 3GPP, ETSI, ITU, GSMA, CAMARA, O-RAN, and TM Forum.
- OSFT Training: Utilizes Orthogonal Subspace Fine-Tuning (OSFT) by Red Hat AI Innovation Team to integrate new domain knowledge without degrading general instruction-following abilities.
- Multimodal Architecture (Text-Only Training): Architecturally supports text and image inputs, though only the text path was trained and adapted for telecom. The vision encoder remains stock Gemma 4.
- Enhanced Data Preparation: Includes broader data mixtures for direct telecom QnA, abstention, RAG, base-model-style telecom data, and general-purpose instruction-following and tool-calling examples.
Good For
- Retrieval-Augmented Generation (RAG) over telecom standards and technical documentation.
- Standards Interpretation and Summarization for major telecom bodies (3GPP, ETSI, GSMA, etc.).
- Product Development and Network Configuration Assistance.
- Telecom-specific Direct QnA.
- Agentic Workflows requiring domain knowledge, with external tools and validation.
It's important to note that while the model has general-purpose tool-calling capabilities, it does not include telecommunications-specific MCP or tool-calling examples. It is a generative text model and not designed for embedding, retrieval, or reranking tasks.