tharaka-axonect/nemotron-testing

TEXT GENERATIONConcurrent Unit Cost:2Model Size:30BQuant:FP8Context Size:32kPublished:Aug 6, 2026License:nvidia-open-model-licenseArchitecture:Transformer Open Weights Featherless Exclusive Cold

AdaptKey-Nemotron-30b is a 30 billion parameter LoRA fine-tuned version of NVIDIA's Nemotron-3-Nano-30B model, developed by AdaptKey. It is specialized for telecommunications and network engineering applications, trained on over 1.3 million telecom domain examples. This model excels in tasks like network log analysis, structured YAML generation for network configurations, and tabular reasoning over network parameters, achieving a 596 composite benchmark score.

Loading preview...

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.