tharaka-axonect/nemotron-testing

AdaptKey-Nemotron-30b: Telecom-Specialized LLM

AdaptKey-Nemotron-30b is a LoRA fine-tuned variant of NVIDIA's Nemotron-3-Nano-30B model, developed by AdaptKey. It is specifically optimized for telecommunications and network engineering tasks, having been trained on over 1.3 million telecom-specific examples covering 3GPP standards, IETF protocols, network traces, and configuration data.

Key Capabilities and Performance

This 30 billion parameter model demonstrates significant improvements in telecom-related benchmarks. It achieved a composite score of 596, representing a +58 point (+10.8%) improvement over the NVIDIA Nemotron-3-Nano-30B-A3B baseline (538). Notable gains include:

• TeleYaml: +16.8 points (+26.9%) for structured YAML generation for network configurations.
• TeLogs: +12.8 points (+26.2%) for network log analysis and fault diagnosis.
• TeleTables: +11.8 points (+19.3%) for tabular reasoning over network parameters.

The fine-tuning process utilized conservative LoRA hyperparameters and a learning rate of 5e-5 to prevent catastrophic forgetting, thereby preserving general language capabilities while enhancing domain-specific expertise.

When to Use This Model

• Telecommunications Engineering: Ideal for applications requiring deep understanding of 3GPP, IETF, ITU, and TM Forum standards.
• Network Operations: Excellent for tasks such as network log analysis, anomaly detection, and troubleshooting.
• Configuration Management: Highly effective for generating and validating structured network configurations (e.g., Open5GS YAML).
• Research & Development: Useful for exploring advanced network slicing, network function configuration, and traffic prediction within telecom domains.